标签： AI引用准确性_en

What Should We Do If AI Answers Contain Negative Information About Our Brand?

After AI search gradually became a primary entry point for users to get information, brand reputation risk has taken on a new form: negative information no longer stays only on web pages, posts, or search results—it gets “summarized” into a sentence or a conclusion and appears directly in AI answers.
The impact on companies is often faster and more concentrated: users may not click links to verify, but they will remember AI output as a “conclusion.”

From the perspective of GEO (Generative Engine Optimization), this article provides an actionable handling process and AI search optimization methods to help you, without departing from facts and compliance boundaries, turn “negative AI answers” into a controllable problem that can be corrected, diluted, and governed over time.

Let’s Unify the Concepts First: What Exactly Do AI Search, AI Search Optimization, and GEO Solve?

What Is AI Search?

AI search does not just “give you a pile of links”; it synthesizes web information, knowledge bases, user content (UGC), news reports, and more into a more “answer-like” result. It may:

cite external pages as evidence;
combine training information already inside the model;
“compress and summarize” content from multiple sources.

This is also why, once negative information is “compressed into a conclusion,” its spread and harm are clearly amplified.

What Is AI Search Optimization?

AI search optimization focuses on: how to make AI, when answering questions related to you, more easily retrieve and understand you, cite you, and output accurate, complete, verifiable conclusions. It is not just keyword rankings but competition over “answer quality and citation quality.”

What Is GEO (Generative Engine Optimization)?

GEO (Generative Engine Optimization) can be understood as content and brand entity optimization for generative engines (LLMs/AI search). Its goal is to make your brand information appear in AI output as:

verifiable (sourced and citable)
understandable (clear structure, explicit definitions)
repeatable (concise conclusions + evidence chains)
sustainable (stable over time and updatable)

Why Does AI Produce Negative Information About Your Brand? (Four Most Common Causes)

To solve “what AI said,” you first need to understand “why AI said it.”

1) AI cites external sources (news, forums, reviews, social media)

AI may retrieve and cite:

news reports (including old news)
forums/post bars/Q&A sites
negative reviews or complaints on review platforms
secondary spread on social media, even rumors

If you can locate the specific source, you can reduce the problem from “abstract AI” to “a specific piece of content or a group of content.”

2) AI amplifies an individual incident into a general conclusion

For example:

one customer dispute → “this company has terrible service”
one forum discussion → “they are going bankrupt”
a piece of outdated information → “they have ceased operations”

A common risk of generative answers is “over-generalization,” which is a typical GEO challenge: AI summarizes well but is insensitive to boundaries and timeliness.

3) Information is outdated but still repeated (update lag)

Even if you have clarified or remediated:

old content may still be indexed and cited;
new content, if not authoritative or clear enough, may not be preferentially adopted by AI;
platform update cadences and crawl cadences differ, so “you have corrected it, but AI still outputs the old conclusion.”

4) The brand lacks “fact anchors,” forcing AI to piece together third-party material

When your official site does not have:

a clear brand introduction (About/company information)
citable fact pages
authoritative, verifiable update notes
citable FAQ/statement pages

AI will rely more on third-party sources, where negative content is more likely to be “narrative and spreadable.”

GEO Challenges and Misconceptions: Why Do Many Companies Become Less Effective the More They Handle It?

The following are high-frequency misconceptions that cause you to invest a lot of effort while AI output does not change—or even worsens.

Misconception 1: Focus only on traditional SEO rankings and ignore the “AI citation chain”

Traditional SEO emphasizes “rankings and traffic,” while AI search emphasizes “citations and repeatable conclusions.”
Even if you rank well, AI may not cite you if the page lacks clear conclusions and evidence chains.

Misconception 2: Only delete posts without building “official citable fact sources”

Deletion or appeals are indeed important, but what matters more is completing citable official information.
Otherwise, even if one negative source disappears, AI may still piece together similar conclusions from other sources.

Misconception 3: Use “emotional rebuttal” instead of “factual clarification”

AI more easily absorbs structured, verifiable content with clear wording.
Emotional, aggressive rebuttals without evidence neither improve credibility nor help AI citation.

Misconception 4: Turn GEO into “keyword stuffing”

In generative engines, “repeating keywords” does not naturally become a “trustworthy answer.”
GEO values more:

whether definitions are clear
whether evidence is credible
whether the structure is easy to cite
whether entity information is consistent (company name, aliases, product names, responsible persons, etc.)

Misconception 5: Ignore “positive clarification of negative keywords”

Many companies are unwilling to mention words such as “bankruptcy/absconding/complaints” on their official sites, but the reality is:
users will ask, and AI will answer.
If you do not provide clear clarification, AI can only cite third-party narratives.

An Actionable Handling Process: Turn “Negative AI Answers” into a Solvable Task List

Below is a GEO handling loop for corporate teams. You can treat it as a “reputation emergency plan for the AI search era.”

Step 1: Reproduce and Collect Evidence (Turn the Problem from “Feeling” into “Evidence”)

You need to fully record:

which AI product/platform was used
how the question was asked (Prompt)
the answer content (original text)
whether there were citation links/sources
time and region (if they differ)
differences across phrasings of the same question

Why does this matter?
Because AI search answers may change with phrasing. You need to lock down “the phrasing that triggers the negative conclusion” before you can fix it in a targeted way.

It is recommended to build an internal table with fields:

question type (bankruptcy/complaint/true-false/competitor comparison, etc.)
negative conclusion summary
cited sources (URL/platform/account)
severity level
owner
handling status

Step 2: Identify the Information Source (Locate “Where the Negative AI Statement Comes From”)

Split the negative answer into two parts:

conclusion: what AI said (for example, “was once complained about for quality issues”)
basis: why AI said it (which sources it cited or what content it inferred)

Common source-tracing methods (by priority):

first check whether AI provides citation links
use key sentences to locate the original text via search engines/site search
check high-authority third parties (media, forums, encyclopedias, review platforms)
check the “secondary spread chain” (reposts, screenshots, copies)

The goal is not “debating AI” but finding “governable sources.”

Step 3: Classify and Match the Handling Strategy (Fact/Misunderstanding/Rumor/Outdated)

Classifying negative information into 4 types greatly improves handling efficiency:

Type	Characteristics	Your goal	Core action
Factual negative	genuinely happened, with evidence	explain and improve rather than deny	public statement + remediation evidence + customer success cases
Misunderstanding negative	partly true but exaggerated/generalized	correct boundaries and semantics	fact clarification page + FAQ + unified external messaging
Rumor/defamation	no evidence, unreliable sources	clean up sources + build authoritative counter-evidence	appeal/legal letter (if necessary) + authoritative statement page
Outdated information	old incident or old status	update timeliness and current status	update note + fact anchor “as of a certain date”

Step 4: Content Response (Publish Formal Clarification in an “AI-Citable Way”)

In the GEO context, “responding” is not posting a social media update but creating retrievable, citable, long-term updatable page assets.

Two types of pages are recommended:

A. “Fact Statement / Clarification” page (for a single rumor)

Page titles should directly match user questions:

“Fact Statement Regarding ‘X’”
“Clarification and Information Update on the ‘X Rumor’”

Suggested content structure (AI-citation friendly):

one-sentence conclusion: clear stance (for example: the rumor is false/remediated/resolved)
fact timeline: what happened, when, and how it was handled
evidence and citations: publicly shareable materials (announcements, reports, media coverage, screenshot descriptions, etc.)
current status: the latest business/operations/qualification status as of a certain date
external contact channels: media contact/customer service/official email
update record: record the date and content of each update

B. “Brand Facts” page (long-term asset)

This is one of your most important “authority anchors” in the AI search era. It does not chase hot topics; it only answers “factual questions AI will be repeatedly asked.”

Example modules:

company/brand introduction (one paragraph + 3 key points)
operating status (as of date)
core products and target audiences
compliance qualifications and certifications (if any)
customers and cases (the publicly shareable part)
clarification of common misunderstandings (fact-based responses including keywords such as bankruptcy/absconding/complaints)
media coverage and newsroom entry
official channel list (official site, WeChat official account, video account, etc.)

Step 5: Optimize the Brand Introduction (the About Page Must Be “Directly Citable by AI”)

You need to upgrade the About page from a “brand story” to a “brand entity specification.”

Suggested writing (better for AI search optimization):

one-paragraph definition: who we are + what problem we solve for whom
three facts: founding time/operating status/service scope, etc. (provide “as of date” where possible)
key data (use exaggeration cautiously; make it verifiable where possible)
clarification of common doubts (turn negative keywords into factual statements)
For example: “As of 2025, the team continues to operate and provide services to X type of customers” (please use your actual latest date here).
authority endorsement entry: newsroom/white papers/cases/media coverage pages

This is essentially GEO: letting AI get “the most repeatable answer” from “the most authoritative source.”

Step 6: Contact Platforms and Govern Sources (Correction Requests + Deletion/Appeals)

When negative information is clearly wrong, rumor, or defamation, handling must include “source governance”:

For AI platforms: submit corrections through their “feedback/report error/appeal” channels
When submitting, include:
screenshots of the wrong conclusion
links to authoritative pages with correct facts (your statement page/fact repository)
an explanation of why it is wrong (outdated/miscited/no evidence)
For source platforms (forums, social media, review platforms, etc.): file appeals, complaints, or deletion/correction requests according to platform rules

Note:
your goal is not “no negative content on the internet” but:

fewer misinformation items
more authoritative information
AI more inclined to cite authoritative information and output accurate conclusions

Step 7: Actively Optimize Positive Assets (Counter Negatives with Positives, but Truthfully)

AI output is often a “weighted summary of materials.” When positive, authoritative, verifiable material is sufficient, negative content—even if it exists—is less likely to become the main conclusion.

What you should do is not “buy fake reviews” but build publicly shareable, verifiable content assets, such as:

customer success cases (anonymized or industry-level treatment)
product quality/service process descriptions (transparency reduces misunderstanding)
team/author/expert endorsement pages (enhance credibility)
newsroom and regular updates (build time-series authority)
industry viewpoint content (bind the brand to professional topics)

Step 8: Establish Continuous Monitoring (Extend Public-Opinion Monitoring to AI Output)

In the AI search era, you need to monitor two lines:

Content line: what new content about your brand appears online (especially high-authority citable sources)
Output line: how AI describes you on key questions (accuracy, cited sources, sentiment)

It is recommended to build an “AI question bank” (continuous retesting):

Is the brand bankrupt/absconding?
How is the brand reviewed? Are there complaints?
How does it compare with competitors?
Is it trustworthy/a scam?
How is after-sales service?
Is it suitable for a certain audience?

Fix several phrasings for each question and periodically sample output and citation changes.

WordPress Implementation Checklist: Turn GEO into Executable On-Site Page Engineering

If you publish with WordPress (as your site also does), treat the following as “GEO infrastructure”:

1) Pages you must build (priority high to low)

About/company information page (citable, with fact anchors)
Brand Facts repository
Newsroom/media coverage aggregation page
Fact statement/clarification pages (created per incident)
FAQ (for high-frequency user questions)
Contact us (official channels)
Cases/customer stories (publicly shareable part)

2) Structural writing (make it easier for AI to “grab conclusions”)

titles directly match user questions (do not write only marketing slogans)
provide a “conclusion paragraph” at the start of each page (1–3 sentences)
use subheadings, lists, and tables to chunk information
provide “as of dates” for key facts
clearly label external citations and evidence (where publicly shareable)

3) “Citation-Friendly Paragraph” Template for AI Search Optimization

You can insert similar paragraphs on key pages (replace with actual facts):

Conclusion: the statement about “X” is inaccurate.
Facts: as of [date], we continue to operate and provide [services/products] to [customer type].
Basis: see the “Update Record” and “Evidence Materials” sections of this page for related notes and updates.
To verify: further information is available through [official contact].

This writing is clear for human readers and easier for AI to extract as a “repeatable answer.”

Practical Examples: How to Handle High-Risk Negative Keywords Such as “Bankruptcy/Absconding/Complaints”?

Scenario A: AI says you have “gone bankrupt”

Possible causes:

old posts/old news
a company address change being misread
short-term business adjustments being exaggerated

Recommended actions:

add “operating status (as of date)” to the Brand Facts repository and About page
publish an “Operating Status Statement/Information Update” page (with contact details)
if there is a source rumor, initiate deletion/correction appeals
use the newsroom/announcements to create continuous updates and build “time-series credibility”

Scenario B: AI says you have “received many complaints”

Possible causes:

concentrated negative reviews on one platform
individual cases being secondarily spread
competitor comparison content steering

Recommended actions:

first distinguish “factual complaints” from “exaggerated misunderstanding”
for genuine issues: publicly disclose improvement measures (process, refunds, after-sales)
for misunderstandings: publish a fact clarification page emphasizing boundaries and data sources
long term: build cases, product transparency pages, and service standard pages to dilute single-point negatives

One-Page Action Checklist (for Internal Team Alignment)

[ ] Reproduce and record the negative AI answer (platform, prompt, original text, citations, time)
[ ] Trace sources: locate the source content and spread chain behind the negative conclusion
[ ] Classify: fact/misunderstanding/rumor/outdated → match the handling strategy
[ ] Build pages: fact statement/clarification (citable structure + evidence chain + update records)
[ ] Strengthen About: add fact anchors and fact-based responses to common doubts
[ ] Build a “brand fact repository”: form a long-term authoritative information source
[ ] Platform corrections: submit corrections/appeals to AI platforms and source platforms
[ ] Positive assets: continuously update the newsroom, cases, transparency pages, and FAQs
[ ] Monitoring: build an AI question bank and periodically retest output and citation changes

Conclusion: AI Is Not “Deliberately Smearing You”—It Is Merely Retelling the Material It Can Get

It is not uncommon for AI to output negative information, especially when third-party content is easier to retrieve and more narratively compelling, while your official information lacks a “citable structure.”
From the perspective of generative engine optimization (GEO), the core of handling such issues is not “arguing with AI” but:

making authoritative facts more visible (retrievable)
making facts easier to cite (extractable)
keeping facts continuously updated (traceable)
governing the sources of errors (cleanable)

Only when you change the “information supply” will AI’s conclusions change accordingly. A company’s PR and content strategy must also incorporate AI search as a new information channel into systematic management—this is now a basic capability for brand growth and risk control.

2026年10月24日

How Can You Verify AI Citation Accuracy in the AI Search Era? GEO Review Processes and Measurement Metrics

Goal: upgrade from “being mentioned by AI” to “being correctly cited, correctly attributed, and not misused, and to be measurable and continuously iterable.”

What you really need to solve is not “whether AI will cite you” but “whether citations are reliable.”

In traditional search, users see a list of links, and the right to interpret information mainly lies with users; in AI search (generative answers/conversational search), users often read the “answer” directly and may not even click links.

This leads to a key change:

AI’s “retelling/summary” is often treated by users as “the fact itself.”
Once quote-mining, lost conditions, or attribution errors occur, the impact is not just traffic but brand credibility and business risk.
Therefore, in generative engine optimization (GEO), “citable” is only the starting point; “correctly citable, verifiable, correctable, and measurable” is the complete loop.

Why Do AI Citations Go Wrong? First Understand the Typical Forms of “Misuse”

When doing AI search optimization, you first need to unify the team’s definition of “misuse.” The six most common types are:

1) Quote-mining: key conditions are deleted

Your original may say:

“Under condition A, X can be done, but situation B does not apply.”

AI, in order to simplify, may turn it into:

“X can be done.”

Result: users act on it, deviations occur, and you take the blame.

2) Over-generalization: treating an “example” as a “conclusion”

The original is a case or partial experience, and AI abstracts it into a universal rule.

3) Misattribution: attribution errors (model “hallucination”)

AI may attribute others’ views to you, or your views to others—a classic “hallucination/attribution confusion.”

4) Multi-source mixing: facts are contaminated after splicing

AI mixes passages from multiple pages, causing your content and others’ content to “synthesize” a new conclusion.

5) Timeliness errors: old versions treated as new conclusions

Your page was updated, but AI still cites the old version’s wording; or cites numbers without date context.

6) Misuse in high-risk fields: medical, legal, tax/finance, etc., treated as advice

Educational content is treated as diagnosis/prescription; compliance notes are treated as executable plans. Such misuse often directly brings complaints and reputational damage.

The Core of GEO: Build a Content and Operations System for “Verifiable Citations”

Many teams only focus on “exposure/mentions” when doing GEO, but what really affects long-term growth are these four outcomes:

can be found by AI (retrievable)
can be cited by AI (citable)
does not easily deform when cited (repeatable without distortion)
when errors occur, they can be found, corrected, and reviewed (governable and measurable)

Below is an actionable process: from review to correction, and then to GEO results and measurement.

An Actionable “AI Citation Accuracy Review Process” (Essential for GEO Operations)

This part is the “routine action” you need to establish; it can run weekly or after publishing important content.

Step 0: Prepare the “Content Baseline” First (Otherwise You Cannot Review)

Build a “baseline checklist” for each important article, preferably at the end of the article (or in an internal document):

this article’s 3–7 key conclusions (one-sentence versions)
the applicability conditions/boundaries corresponding to each conclusion (must be retained)
the version and date
of key numbers/definitions

approved standard phrasing for citation (how you want AI to say it)

The point of this step: make review no longer rely on “feelings” but on explicit comparison.

Step 1: Check AI Output (Against the Original, Catch “Lost Conditions”)

Compare the AI search answer sentence by sentence with the original, focusing on:
whether rewriting changes the meaning (especially strong tones like “can/must/recommend/certainly”)
whether limiting conditions are lost (audience, scenario, time, premise)
whether “possible” becomes “certain”
whether an “example” is treated as a “conclusion”

whether a “concept explanation” is treated as an “action instruction”

Practical advice (plain but effective): three-color labeling
Green: accurate retelling with complete conditions
Yellow: conclusions broadly correct but conditions/scope missing

Red: wrong conclusion, wrong attribution, or risky misleading

Step 2: Focus Specifically on “Attribution Errors” (a High-Frequency Zone for Model “Hallucination”)

You need to check two things separately: whether your brand/website is explicitly mentioned
(e.g., “UME / growume.com”) whether attribution is correct when mentioned

(whether others’ views are attributed to you)

Once an attribution error is found, handle it by priority: High priority
: medical/legal/tax-finance; operational advice that may cause losses; obvious misleading Medium priority
: key strategies distorted; conclusions directionally wrong Low priority

: wording not rigorous enough but not affecting understanding

Step 3: Risk-Downgrade “Serious Information” (Medical/Legal/Tax-Finance, etc.)

If your content involves risk fields, design two layers of anti-misuse protection in the content structure:

(1) Upfront disclosure: tell AI and users “this is not advice”
applicable scope: education/experience sharing/framework methods
non-applicable scope: individual diagnosis, legal opinions, tax filing decisions

mandatory action: consult a professional for specific situations
(2) Downstream context: tell AI “what the conditions are”

after the conclusion paragraph, add:
applicable conditions
exceptions
risk reminders

a checklist requiring further verification (so AI dares not fill gaps)

You are not “done by adding a disclaimer”; you need to make the content itself more “immune” to misuse.

Step 4: Monitor User Feedback (Complaints, Comments, and Tickets Are All “Misuse Signals”)

If users have a worse experience because AI cited your content, you will usually see:
comments questioning: “ChatGPT said according to your article X can be done, but the result was wrong”
DMs/ticket complaints: your brand was treated as an endorsement

social media discussion: screenshots spreading wrong understanding

This feedback is critical in GEO because it shows:
AI’s wrong retelling has “entered the communication chain”

your content may have expression structures that can be misread Build a “misuse feedback table”

, with each record containing at least:
triggering platform (which AI search/which scenario)
the user’s original words (screenshot/text)
the corresponding URL of your page
error type (lost conditions/wrong attribution/splicing contamination/timeliness issue)
impact level (high/medium/low)

handling method and result

Step 5: Establish GEO Content Quality Standards (Reduce Misuse Probability at the Source)

You can create a “GEO content quality checklist” for the team. Here is a directly reusable version:	Check item (writing/structure)	What you need to ensure
Why it reduces AI misuse	Conclusion first	each section starts with a one-sentence conclusion
AI can grab the correct main idea more easily	Complete conditions	follow the conclusion with applicable/non-applicable scope
prevent quote-mining	Clear definitions	core terms have fixed definitions and examples
prevent concepts being replaced/misunderstood	Strong/weak tone standards	possible/usually/under…circumstances/not guaranteed
prevent AI from forcing “certainty”	Numbers have timestamps	data, ratios, and checklists are dated
prevent timeliness errors	Counterexamples/boundary notes	explicitly list “non-applicable” situations
AI dares not generalize	Citable passages	key conclusions use short sentences, lists, and steps
facilitate correct AI extraction	Versions and changelog	important content shows “updated on” and change points

facilitate correction and external explanation

Writing That Makes Content “More Suitable for AI Search Citation” (Content Engineering for AI Search Optimization) Below is a content design method more oriented toward “generative engine optimization (GEO),” with the core goal: make it harder for AI to extract incorrectly

.

1) Use the three-part expression “conclusion + conditions + exceptions”

Recommended template: Conclusion:
I recommend X Conditions:
only valid under A/B/C Exceptions:

if D/E occurs, do not use X; use Y instead

This fits AI search’s extraction logic better than “long paragraphs of exposition.”

2) Write “key points” as citable “atomic sentences”

Atomic sentence characteristics:
a single sentence expresses the complete meaning
does not depend on the previous paragraph to be understood

avoid unclear references (“it,” “this,” “the above”)

AI is better at citing “independent and complete” sentences.

3) Add “guardrail paragraphs” to high-risk content

You can add a fixed module in the article, for example:
Risk notice (must read):

1) This section is for knowledge education only; 2) it does not constitute individual advice; 3) verify conditions and consult a professional before acting.

Such structured guardrails can significantly reduce the consequences of misuse (even if AI may still err, users are more likely to be reminded).

4) Use “common misreadings” to inversely train AI

Add a short section at the end of the article: Common misreading 1:
treating A as B (why it is wrong) Common misreading 2:

ignoring condition C (what it will cause)

This not only serves users but also gives AI stronger semantic boundaries.

5) Use a “changelog” to counter timeliness errors

At minimum, add to important pages:
Updated on: YYYY-MM-DD

This update: which conclusions or conditions were added/corrected

When AI search cites an old conclusion, it is also easier for you to explain externally.

After Finding an AI Citation Error, How Should You Correct and Govern It?When you find a wrong citation in AI search, do not just “get angry”; treat it as part of GEO operations: discover—attribute—fix—verify—consolidate

.

1) Content-side fixes: make the original less likely to be misread

Prioritize fixing these three types of expression:
conclusions buried in long paragraphs (conditions are easily lost during extraction)
key conditions written separately (AI takes only the conclusion)

tone too strong (easily treated as certain advice)

The fix is usually simple: rewrite key conclusions in the “three-part” form and keep conditions close to conclusions.

2) On-site fixes: publish a “clarification/correction note”

If the error has already spread, build a citable correction entry on the website (a separate page or an in-article module):
the wrong statement (AI output)
the correct statement (your original meaning)
the conditions that a correct citation should include

a link to the original

Such pages often have extra value in GEO: when AI retrieves related questions again, it is more likely to find the “corrected version.”

3) Platform-side feedback: report the issue to the corresponding channel

Feedback entries differ across platforms, but your principles should be consistent:
provide verifiable evidence: AI output screenshot + your original URL + point out which sentence is wrong
state the risk level: whether it may cause misleading/loss

state expected fixes: correct the citation, add links, correct attribution

The focus is not “complaining” but “providing reproducible evidence and correction text.”

GEO Results and Measurement: How to Quantify “Whether AI Citations Bring Growth,” and Keep It Controllable By now you have review and correction capabilities. The next step is GEO results and measurement

: turn “citations” into a metric system, otherwise you cannot explain input-output to the team/boss.

1. Use “Five Layers of Metrics” to Measure GEO Visibility layer
: whether you appear in AI search Citation layer
: whether your pages/viewpoints are cited Accuracy layer
: whether citations are correct and conditions retained Action layer
: whether clicks, inquiries, and conversions are generated Risk layer

: whether misuse, complaints, and correction costs occur

2. Core Metric Definitions (Ready for a Weekly Report)

Here is a set of metrics that are both “actionable” and “explainable”:	Metric	Definition
The problem you are solving	AI visibility rate	the proportion of sampled questions where your brand/page appears
whether you enter the AI answer ecosystem	Citation rate	the proportion of sampled questions where AI explicitly cites your content/link
whether you are treated as an information source	Attribution accuracy	the proportion of citations with correct brand/author/website attribution
whether misattribution occurs	Citation accuracy	the proportion where key conclusions + key conditions + key numbers are all correct
whether you are “retold correctly”	Condition retention rate	the proportion of AI outputs that retain the original’s applicability conditions
whether quote-mining is likely	Misuse event rate	the number of misuse/complaints/corrections over a period
whether risk is controllable	AI-referred traffic share	the share of visits from AI search-related sources
whether traffic is generated	AI-referred conversion rate	sign-ups/inquiries/deals converted from AI referrals

whether business results are generated

Note: do not look only at “mention volume.” In GEO, “wrong mention” can be worse than “no mention.”

3. How to Collect Data? Combine “Sampling Audit + Behavioral Data”
(1) Sampling audit (for accuracy and attribution)

Sample 10–30 questions strongly related to your business each week (covering brand terms, product terms, and question terms), test in multiple AI search scenarios, and record:
AI output text
whether cited/linked
comparison of cited passage with the original
red-yellow-green labeling

error type and risk level
(2) Behavioral data (for growth and conversion)

Track with your existing analytics system (e.g., logs, GA4, lead system):
whether AI platforms appear in source/referral
whether landing pages are “citable content pages”

whether the conversion path is shortened (users inquire/order faster)

4. Give the Team a “GEO Citation Health” Score (for Reporting)

You can use a simple score to explain a complex situation clearly:
GEO citation health =
40% * citation accuracy +
30% * attribution accuracy +
20% * condition retention rate +

10% * misuse events (reverse deduction)

This lets you answer three things in the weekly report:
Did AI search performance improve this week?
Which pages are most likely to be misused?

What are the next optimization priorities?

One-Page Quick Check List (For Pre-Publish/Retrospective)

Before publishing or updating an important piece, self-check with this list:
[ ] The title and opening give a clear conclusion within the first 100 words
[ ] Every key conclusion is immediately followed by “applicable/non-applicable scope”
[ ] Key terms have definitions (ideally with examples/counterexamples)
[ ] Strong tones (must/certainly/guarantee) have evidence or are changed to conditional sentences
[ ] Data and checklists are marked with dates or versions
[ ] The article ends with “common misreadings/boundary notes”
[ ] There is an update date and changelog (mandatory for important pages)
[ ] Content on medical/legal/tax-finance topics has explicit disclaimers and risk notices
[ ] Select 3–5 typical questions and self-test whether AI search retells them accurately

[ ] Establish a sampling audit record (to provide a baseline for GEO results and measurement)

Conclusion: AI Citation Errors Cannot Be Completely Avoided, but They Can Be “Designed, Reviewed, Governed, and Measured”

In the AI search era, GEO competitiveness comes not just from “writing more” but from:
content being understood and cited by generative engines
citations not easily misread or misused
errors being quickly corrected and spread reduced

the whole process being measurable, iterable, and reviewable When you get this process running, your generative engine optimization (GEO)

2026年10月24日

GEO Effectiveness and Measurement: A Complete Metric System from AI Visibility to Conversion and ROI (with Templates and FAQ)

After AI search became a mainstream entry point, many enterprises and content teams encounter the same real dilemma: GEO (Generative Engine Optimization) seems “effective,” but it does not necessarily show up as “traffic growth” on the data dashboard. The reason is not complicated — generative engines prefer “giving answers directly” rather than “giving links for you to click.” So a typical phenomenon appears: content is frequently cited, summarized, and recommended by AI, and users act on it, but clicks do not necessarily happen, and sometimes there are no links at all.

This does not mean GEO is unmeasurable. Quite the opposite: the focus of GEO measurement needs to migrate from “clicks” to “visibility and influence at the answer layer,” then connect influence to subsequent behavior and final conversion in an explainable chain.

This article uses a finer, more accessible, and practical framework to break down GEO effectiveness, build the metrics, create the dashboard, and give you templates you can reuse directly (question bank tables, visibility scoring, attribution and ROI ranges, competitor benchmarking, accuracy audits, and review cadence).

1. Align First: GEO’s “Effectiveness” Is Not Equal to “Traffic”

The traditional SEO funnel is usually:

Impression → Click → Conversion

But in the AI search context, a more realistic funnel is:

Exposure (AI visibility) → Guidance (subsequent behavior) → Conversion (business results)

The biggest change is: “click” is no longer a mandatory step. Users may complete understanding, comparison, and decisions within an AI answer, and only then search for the brand name, type in the URL directly, consult sales, or search on e-commerce platforms… So if you stare only at clicks and sessions, it is very easy to misjudge GEO as “ineffective.”

2. GEO’s Three-Layer Effectiveness Model: Exposure, Guidance, Conversion (and What to Look at in Each Layer)

Making GEO “measurable” is essentially: systematizing the three layers of metrics and connecting them in the same dashboard.

2.1 Layer One: Exposure (AI Visibility) — Are You in the AI Answer? How Do You Appear?

The questions you need to answer:

Does AI mention you when answering related questions?
Is it a “passing mention” or “a key recommendation/main source”?
Is AI’s restatement accurate, and does it match your positioning and selling points?

Common metrics (from easy to hard):

Mention Rate
Citation Rate
Link Rate
Query Coverage
Answer Share / SOV
Accuracy / Consistency

Plain understanding: if users frequently “see your name/viewpoints” without clicking links, that is already the first layer of GEO success.

2.2 Layer Two: Guidance (User Behavior) — Even Without a Link, AI Can Push Users to Act

The questions you need to answer:

Does AI’s answer prompt users to search for you further, visit you, or consult you?
Is there a trend change in this “dark-path” behavior?

Common metrics:

Branded Search Lift
Direct / Returning visits increase
Changes in on-site search terms (brand terms, brand + category terms)
Frequency of “mentioning AI” in consultations/forms/customer service
Increase in content redistribution/external mentions

Plain understanding: when you find that brand-term search, direct visits, and the proportion of user consultations mentioning AI are rising, GEO is starting to work at the “guidance” layer.

2.3 Layer Three: Conversion (Business Results) — Ultimately Answering “Did It Bring Money/Leads/Orders?”

GEO conversion is often “indirect”: users are first seeded by AI, then search, then visit, then compare, and only then convert. So you need more reasonable attribution methods to “see” it.

Common methods (recommended to do at least 2–3):

Add source self-reporting to forms/sign-ups/consultations (including options for “AI search/ChatGPT/LLM recommendation”)
Standardize sales/customer service scripts and put “from AI or not” into CRM fields
Design AI-exclusive resources/landing pages/copyable links to improve trackability
Use experiments/controls to observe incrementality (strengthen GEO for one product line/topic, leave the control group unchanged)

3. An “Executable” GEO Measurement System: Three Types of Metrics + One Dashboard

You do not need to make it complex from the start. It is recommended to begin with a minimum viable system (MVP) that “can run,” then upgrade to a standard/scaled version.

3.1 GEO Dashboard (Suggested Structure)

Layer	The question you need to answer	Recommended metrics (examples)	Common data sources	Frequency
Exposure (AI visibility)	Am I in the AI answer? How do I appear?	Mention rate/citation rate/link rate, coverage, Top-3 share, accuracy score, AI-SOV	Fixed question bank sampling, manual review	Weekly/biweekly
Guidance (user behavior)	Does AI push users to act?	Branded search trends, direct/none deep landing pages, return visits, on-site search, “from AI” share in surveys	GSC/GA4/on-site search/surveys	Monthly (trends can be weekly)
Conversion (business results)	Does it bring leads/orders/revenue?	AI-influenced leads, AI-assisted conversion rate, AI-influenced revenue, CAC changes	CRM/forms/order systems	Monthly/quarterly

The key value of such a “layered dashboard” is: you can explain “why there was no obvious growth this month,” and also point out “upstream signals are improving, and what to do next.”

4. Step One: Set the Ruler First — Build an AI Search Question Bank (Query Set)

In GEO, what you really need to measure is not a page but the questions users will ask (Query). Therefore building a “fixed question bank” is the foundation of the entire measurement system.

4.1 How to Build a Question Bank That Is “Sufficient and Comparable”?

It is recommended to start with 50–200 entries, covering real user intents and classified by decision stage:

Cognitive stage: what it is/why/principles/trends
Comparison stage: how to choose/comparisons/recommendations/Top options
Decision stage: pricing/cases/implementation/risks/alternatives

And set a business weight for each question (e.g., 3/2/1):

3 points: strong commercial intent (selection, quotes, vendor recommendations)
2 points: medium intent (methods, steps, best practices)
1 point: general awareness (concepts, trends)

4.2 Measurement Frequency and Sample Size Suggestions (Avoid Being Misled by “Random Fluctuation”)

A commonly used and pragmatic tiering:

Starter version: 50 questions × 2 platforms × once per month
Standard version: 100–200 questions × 3–5 platforms × once per month (key topics can be weekly)
Scaled version: tool/API automated monitoring + manual spot checks

For comparability, it is strongly recommended to:

Fix the question bank: track the same set of questions over the long term
Fix the platform: keep the entry point consistent (do not mix “with retrieval” and “without retrieval” in the same basis)
Record version and context: different model versions/contexts affect results

5. Step Two: Measure AI Visibility — from “Mention Rate” to a “0–100 Index”

AI visibility is not “how much traffic you have” but: the probability and intensity with which AI mentions/cites/recommends you in a set of AI Q&A scenarios related to your business.

5.1 The Three Most Basic and Universal Metrics

Break “appearance” into three levels (from weak to strong):

Mention: the brand/product name appears, but the source is not explicit
Citation: you are explicitly treated as an information source (viewpoint/data/method)
Link: a traceable link is provided (pointing to your site/page)

In a fixed question bank, you can calculate directly:

Mention rate = number of mentions / total questions
Citation rate = number of citations / total questions
Link rate = number of links / total questions

Example (for business communication):

Of 100 questions, 20 answers mention you → mention rate 20%
Of those, 8 cite you → citation rate 8%
Of those, 5 provide links → link rate 5%

5.2 Turning “Visibility” into a 0–100 Index: Why and How?

Enterprise management usually needs metrics that are “comparable, trackable, and KPI-able,” so indexation is practical. But be clear: an index is not “truth,” only a comparison tool after unifying the basis.

A plain and implementable 100-point model (copy it directly):

(1) Base score per Q&A (0–5)

0: not mentioned at all
1: mentioned only
2: mentioned + explains what you do
3: cites your viewpoint/method/data
4: cites + provides a link
5: as the primary recommendation/key solution (with citation or link support)

(2) Position weight (w_pos): the first source is more valuable
(3) Mention-type weight (w_type): subject-level mentions have greater influence
(4) Platform weight (w_platform) and question weight (w_query): aligned with business priority

Final formula (can go directly into a spreadsheet):

Score per question:
Score_i = Base_i × w_pos_i × w_type_i × w_platform_i × w_query_i
Normalize the total to 0–100:
AI Visibility Index = ( Σ Score_i / Σ (5 × 1.0 × 1.0 × w_platform_i × w_query_i) ) × 100

5.3 Record Field Template (Build the Table Directly from This)

Below is a sample of “visibility monitoring table” fields (Excel/Feishu Sheets/Notion all work):

Field	Description	Example
Date	Test date	2026-01-02
Platform	Platform/model entry	ChatGPT / Bing / Claude
Query_ID	Question ID	Q-001
Query	Original question text	“What is the difference between GEO and SEO?”
Intent_Type	Intent type	Education/comparison/execution/procurement
Query_Weight	Weight (1–3)	2
Brand_Mention	Mentioned (0/1)	1
Citation	Cited (0/1)	1
Link	Link provided (0/1)	1
Base_Score	Base score (0–5)	4
Position	Citation/recommendation position	1/2/3/NA
w_pos	Position weight	1.0
Mention_Type	Mention type classification	Subject-level/recommendation/list/passing
w_type	Mention-type weight	0.8
Notes	Notes	“Link points to a certain article on the official site”

6. Step Three: Make the “Invisible Influence” Explicit — Cross-Validate with Multiple Signals (Triangulation)

GEO attribution is hard to make as precise as SEO clicks, because many AI scenarios clear the referrer, users copy and paste, or they come back later through brand search. So the more robust approach is: cross-validate with 5–8 types of signals rather than betting on one metric.

Below is a combination set ordered from “low cost to high cost, hard evidence to soft evidence.”

6.1 Method One: Capture AI Referrals (the Lowest-Cost “Hard Evidence”)

Identify referral sessions from AI product domains in GA4.

In GA4 explorations/filters (where regex is supported), you can use a rule like this (add or remove as needed):

(chat.openai.com|perplexity.ai|copilot.microsoft.com|gemini.google.com|bard.google.com|claude.ai|you.com)

Be clear about the limitations too: AI referrals are often “very small,” because the referrer may be downgraded, and links copied by users become direct/none; some scenarios may still appear as google / organic. So it is hard evidence, but not the full picture.

6.2 Method Two: Monitor the “Dark-Path Growth” of direct/none

Create a simple GA4 exploration report:

Filter: Session medium = (none) or Session source = (direct)
Row dimensions: Landing page + query string
Metrics: Sessions, New users, Conversions

Focus on observing:

Which “deep content pages” are seeing direct/none growth?
Does it align with “topics more likely to be cited by AI” (definitions/steps/checklists/comparisons)?

This is not strict attribution, but it is very suitable for trend judgment.

6.3 Method Three: Improve Trackability — Design Controllable Parameters for “AI Sharing/Copying”

You cannot control whether AI includes UTMs, but you can increase the probability that “parameterized links are copied and spread”:

https://www.growume.com/xxx?utm_source=ai&utm_medium=referral&utm_campaign=geo

Practical suggestions:

Provide “citation/sharing dedicated links” (with parameters) on key pages, placed at the end of articles or in citation modules
At the same time, keep the canonical pointing to the parameter-free main URL to avoid SEO duplicate content issues

6.4 Method Four: Branded Search Volume (the Most Realistic Chain in the AI Era)

Many AI influences do not manifest as “clicking links” but as:

Being seeded → searching for the brand name/brand + category terms a few days later → then converting

So you should focus on monitoring:

Brand-term query trends in GSC
Trends of combined terms such as “brand + category term/methodology term”
For domestic businesses, add: Baidu-related tools, on-site search, Xiaohongshu/Zhihu/WeChat mention trends, etc.

6.5 Method Five: User Self-Reported Attribution (Most Direct, Most Controllable, Strongly Recommended to Prioritize)

Add one question at key points such as sign-up/consultation/purchase:

“Where did you first hear about us?”
AI search/ChatGPT/LLM recommendation
Search engine
Friend recommendation
Social media …

This is especially effective for “no-link scenarios”; it is recommended to put it into a CRM field and accumulate it over time as AI-Assisted Leads.

6.6 Method Six: Code/Coupon Tracking (Strong Evidence + Easy Content Localization)

In content structures more likely to be cited by AI (FAQs, step checklists, definition/comparison conclusions), place a natural prompt:

“If you saw this content through AI search, feel free to enter the source code GEO-XXX in the form, and we will send you a resource pack.”

Code tracking can answer:

Whether anyone explicitly “comes from AI”
Which types of pages more easily produce AI citation conversions
What incentives AI users want more (checklists/templates/cases)

7. Step Four: Align GEO with Finance — ROI Evaluation and Range-Based Attribution

Many teams get stuck at “management will not approve the budget,” essentially because GEO has not been turned into a controllable, reviewable business project. Although ROI is hard to calculate daily, it can definitely be advanced in layers of “quantitative + qualitative.”

7.1 Why Is GEO ROI Harder to Calculate Than SEO?

Common objective reasons include:

Longer attribution chain (AI first → brand search → organic/private-domain conversion)
More dark traffic (copy-paste, screenshots, word of mouth)
Inconsistent platform definitions (display and citation mechanisms differ greatly across AI products)
Longer effective period (more like content-asset compounding)
Higher share of intangible benefits (trust and mindshare are hard to monetize directly)

Therefore a more reasonable strategy is:

Use “attributable hard metrics” as the baseplate, and fill in the black-box part with “trend validation + qualitative evidence.”

7.2 The Four-Step ROI Method (Ready for Internal Reviews)

Define the goal (leads, revenue, or brand mindshare — choose one or a combination)
Break down costs (labor, content, technology, tools, distribution, management and compliance)
Break down output

Hard output: leads, revenue, subscriptions, add-to-cart, etc.
Soft value: improvements in mention rate/citation rate/coverage/authority signals, etc.

Attribution and calculation (handle the “calculable” and “non-calculable” parts separately)

7.3 Cost Breakdown Checklist (Recommended Monthly/Quarterly)

Content production costs (topic selection, research, interviews, writing, design)
Content updates and maintenance (“content freshness” matters in the AI era)
GEO-specific optimization (Q&A transformation, evidence strengthening, structured components)
Technology and data costs (tracking, UTM, log analysis, dashboards)
Tool costs (monitoring/research/analysis)
Operations and distribution costs (PR, partnerships, authoritative resource development)
Management and compliance costs (brand standards, review, risk control)

7.4 ROI Calculation Suggestion: Use “Ranges” to Express Uncertainty (Easier to Win Management Trust)

You can produce three tiers of estimates:

Conservative: count only the hardest evidence (AI referral + code/coupon)
Base: add source self-reporting (surveys/forms/CRM)
Optimistic: further discount a portion of direct/none abnormal growth (converted by weight)

This way you are not “guessing that AI brings a lot,” but presenting “layered evidence + range expression.”

8. Step Five: Competitor Benchmarking — Upgrade from “Whether I’m Mentioned” to “Whether Share of Answer Improved”

In the AI search era, “we did GEO” does not equal “we are stronger than competitors.” A more executable approach is: establish a baseline with same-question testing, and benchmark gaps with a 5-dimension framework.

8.1 Five-Dimension Benchmarking Framework (Recommended for Long-Term Use)

Visibility: frequency, position, presentation form (recommendation/comparison/passing mention)
Content Gap: what competitors cover that you do not, or what you cover but cannot be cited
Sentiment & Positioning: how AI describes you and competitors, and whether it deviates from positioning
Evidence & Citations: whom does AI prefer to cite? What types of evidence does it cite?
Coverage: in which topics AI mentions only competitors and not you, and why

8.2 “Same-Question Testing” Prompt Templates (Reduce Randomness)

You can apply the following templates directly (three types: recommendation/comparison/evidence):

Template A: Recommendation type (recommend 3 options + comparison table)
Template B: Comparison type (you vs. competitors 1/2, output pros and cons and choice advice)
Template C: Evidence type (require explaining which public source types the basis comes from; if unavailable, explain why and how to verify)

8.3 The Key to Content Gaps: What You Lack Is Not “Article Quantity” but “Citable Answer Modules”

Content AI is more willing to cite usually has:

Clear definitions (boundaries/applicable conditions explicit)
Step-based structure (processes, checklists, templates)
Evidence-based structure (data, cases, sources)
Reusability (paragraphs/tables/key points can be extracted directly)
Continuous updates (fresh information, clear revision records)

It is recommended to use the “question—answer—evidence” three-column method to inventory gaps and turn conclusions into content-asset building tasks.

9. Step Six: Don’t Just Pursue “Being Mentioned,” Ensure “Being Mentioned Correctly” — AI Citation Accuracy Audit

In AI search, users often treat “the answer” directly as fact. Once there is quote-mining, missing conditions, or attribution errors, what is affected is not just traffic but brand credibility and business risk. So in GEO, “being citable” is only the starting point; “being correctly cited, verifiable, correctable, and measurable” is the closed loop.

9.1 Six Common Types of AI Citation Errors (Define Them as a Team First)

Quote-mining (key conditions deleted)
Over-generalization (treating an example as a conclusion)
Misattribution (wrong attribution/hallucination)
Multi-source blending (splicing pollution)
Freshness errors (old versions treated as new conclusions)
Misuse in high-risk fields (medical/legal/tax, etc., treated as advice)

9.2 Sampling Audit Process: Red-Yellow-Green Labeling (Very Useful)

Sample 10–30 strongly related questions weekly, test them across multiple AI scenarios, compare with the original text sentence by sentence, and focus on:

Whether limiting conditions are lost (audience/scenario/time/premise)
Whether “possible” becomes “certain”
Whether attribution is correct (someone else’s view presented as yours, or vice versa)

Three-color labeling:

Green: accurate restatement, complete conditions
Yellow: broadly correct, but conditions/scope missing
Red: wrong conclusion/wrong attribution/risky misleading

9.3 Give the Team a “Citation Health” Score (for Weekly Reports and Governance)

A simplified model you can reuse directly:

GEO Citation Health =
40% * citation accuracy +
30% * attribution correctness +
20% * condition retention +
10% * misuse events (reverse deduction)

10. Step Seven: Feedback Cycle and Review Cadence — How Long Until Results Are “Normal”?

GEO feedback cycles are strongly related to AI scenarios. A pragmatic conclusion applicable to most industries is:

Short-term visibility (a few days to 2 weeks): more common in AI search “with real-time retrieval” (provided the page is indexed and citable)
Medium-term momentum (about 3 months): more stable citation/mention trends begin to appear, especially when you form topical content clusters and entity consistency
Stage evaluation (6–12 months): better suited for systematic reviews: visibility curve, brand mention rate, whether business leads improved

Also note: different AI products acquire information differently (real-time retrieval vs. training data updates), so before evaluating effectiveness you must first confirm “which AI scenario you are evaluating.”

10.1 Recommended Review Cadence (Not Anxious, but Not Delayed)

Weekly: look at early signals (visibility sampling, AI referral, codes/surveys)
Monthly: look at trends (full question bank retest, brand terms, conversion data)
Quarterly: do upgrades (question bank updates, competitor benchmarking reviews, ROI range rechecks)

11. Step Eight: How to Adjust When Results Are Poor — Use a “Diagnosis Tree,” Don’t Guess

When results are still unsatisfactory after 3–6 months, the two most common mistakes are:

Treating “no improvement” as “no value” (ignoring cycle and attribution-chain differences)
Changing before measuring (no metric system, so the more you change, the messier it gets)

It is recommended to diagnose and adjust in order of priority:

11.1 First Distinguish: Which Layer Is the Problem In?

Visibility layer fails: AI rarely mentions/cites you
Traffic layer fails: AI may mention you but does not bring users (or brings very few)
Conversion layer fails: AI traffic exists but does not convert (path and matching issues)

11.2 The Six Most Common Root Causes and Corresponding Actions (by Priority)

Content homogenization: AI has no reason to choose you

Add differentiated value: original methodology, first-hand data/cases, clear boundaries and risk warnings, reusable materials (tables/checklists/formulas/templates)

Technical crawlability issues: efforts are discounted by “not being crawled”

Check robots/noindex/canonical/redirect chains
Check server logs to confirm whether crawling is blocked (403/5xx, etc.)

Competitor content better matches AI preferences

Reverse-engineer: collect 20–50 sample pages frequently cited by AI, summarize structure, information density, authority signals, update frequency, and citable materials, forming an “AI preference list”

Coverage too narrow: you optimized only 10 questions, but users ask 100

Use a “question map” to manage content (question, intent, content format, URL, priority), with the goal of covering 70–90 of the 100 questions users ask most

Lack of external authority and mentions: AI is more willing to cite “widely recognized” content

Complete the author/team/case system; strengthen external signals with off-site media partnerships, white papers and reports, interviews and podcasts, and digital PR

Not enough time: the method is right, but it has not yet entered compounding

Run iterations on a 30/60/90-day cadence:
- Days 31–60: benchmark competitors to upgrade content and fill gaps
- Days 61–90: introduce authoritative assets and scalable SOPs

12. Step Nine: How to Explain GEO Value to Executives/Cross-Department Teams (So the Budget System Understands)

Management’s doubt that “it is invisible and intangible” is reasonable. The solution is not to argue but to translate GEO into three languages they are familiar with:

Trend language: the entry point is changing; not doing it means losing distribution position
Competition language: competitors are taking positions; not following up means losing mindshare at the “answer layer”
Business language: controllable investment, measurable metrics, reviewable pilots, and inferable ROI

12.1 The Most Effective Approach: a 60–90-Day Pilot + Milestone Metrics

Weeks 0–2: build a baseline (30–50 question bank entries, recording visibility, citation rate, expression accuracy, competitor share)
Weeks 3–6: transform high-ROI pages (product/solution/comparison pages + FAQ knowledge base, with answer block structure and evidence reinforcement)
Weeks 7–10: strengthen trust signals (cases, data definitions, source citations, unified brand statements)
Weeks 10–12: retest and review (look at changes in the order of visibility → citation rate → high-intent visits → conversion contribution)

Have executives personally “search once” for industry selection, solution comparison, and price/risk questions, then present the pilot plan and dashboard; this is usually more effective than explaining concepts.

Conclusion: The GEO Measurement System Is Taking Shape; Iterate with the “Scientific Method”

Just as SEO took years to build a mature metric system, GEO’s measurement system is also rapidly taking shape. You do not need to get it perfect in one step, but you must persist with two things:

Set the ruler: fixed question bank + fixed basis + comparable periodic retests
Assemble the puzzle: visibility (upstream) × behavioral signals (midstream) × conversion and ROI (downstream), cross-validated by multiple signals

When you keenly capture effective signals, validate hypotheses with data, and turn each review into the next round of content and asset-building actions, you can take the initiative in the new game of AI search and make GEO a growth engine that is “reviewable, scalable, and sustainable.”

2026年10月24日