By 
Short answer
A GEO checklist scores how ready a page and brand are to be cited in AI answers. This one has 30 checks worth 100 points across six areas: crawler access, answer structure, evidence, entity clarity, off-site presence, and measurement. Score your 5-10 highest-intent pages, fix anything under 70, and re-audit quarterly.
Key takeaways
- Start with access: OpenAI excludes sites that block OAI-SearchBot from ChatGPT search answers, and Google requires pages to be indexed and snippet-eligible for AI Overviews and AI Mode.
- Evidence is the highest-yield on-page fix: in the GEO study, adding quotations, statistics, and cited sources lifted visibility in AI answers by about 28-41%.
- Off-site presence carries 15 of the 100 points because engines lean on reviews, comparisons, and discussions such as Reddit for evaluative questions.
- Score pages that answer buying questions first: comparison, alternatives, pricing, and category pages, not your blog archive.
- A score of 85 or more means the page is citation-ready; under 50 means fix access and structure before anything else.
In this guide
- How do you use this GEO checklist?
- A. Is the page reachable by every AI engine? (20 points)
- B. Does the page answer questions in liftable passages? (25 points)
- C. Does the page give engines evidence to quote? (20 points)
- D. Do engines know exactly who you are? (10 points)
- E. Do trusted third parties mention you? (15 points)
- F. Are you measuring the outcome? (10 points)
- What does your GEO score mean?
- A worked example: scoring a comparison page
- How do you run this audit faster?
Most GEO advice is a list of principles. Principles do not tell you whether a specific page is ready to be cited, or what to fix first. A checklist with points and pass/fail tests does.
This audit has 30 checks worth 100 points across six areas. Each check includes how to test it. The weights are Sleeng's editorial judgment, informed by the research and platform documentation linked throughout, not a published standard. Adjust them if your category behaves differently.
How do you use this GEO checklist?#
Pick 5-10 pages that answer your buyers' highest-intent questions, score each page against every check, and fix the lowest-scoring area first. Brand-level checks (entity clarity, off-site presence, measurement) are scored once and applied to every page.
Scoring rules:
- Award full points only when the check fully passes.
- Award half points for partial passes where noted.
- Score honestly. A generous audit just postpones the fix.
| Area | Checks | Points | Scored per |
|---|---|---|---|
| A. Crawler access and indexing | 6 | 20 | Page |
| B. Answer structure | 6 | 25 | Page |
| C. Evidence | 5 | 20 | Page |
| D. Entity clarity | 4 | 10 | Brand |
| E. Off-site presence | 4 | 15 | Brand |
| F. Measurement | 5 | 10 | Brand |
| Total | 30 | 100 |
A. Is the page reachable by every AI engine? (20 points)#
A page has to be crawled and indexed before any engine can cite it. Google's AI features documentation says a page must be indexed and eligible for a snippet to appear as a supporting link in AI Overviews or AI Mode, and each AI company documents its own crawlers.
| # | Check | Points | How to test |
|---|---|---|---|
| A1 | Page returns HTTP 200 and Googlebot and Bingbot are allowed | 3 | curl -I https://yoursite.com/page and read robots.txt |
| A2 | AI search bots are allowed: OAI-SearchBot, Claude-SearchBot, PerplexityBot | 4 | Read robots.txt and any CDN or firewall bot rules |
| A3 | Page is indexed in Google and not restricted by nosnippet or a very low max-snippet | 3 | Search Console URL Inspection |
| A4 | The main answer appears in the raw HTML, not only after JavaScript runs | 4 | curl -s https://yoursite.com/page | grep "a sentence from your main answer" |
| A5 | Canonical points to itself and there is no stray noindex | 3 | View source, check rel="canonical" and robots meta |
| A6 | A visible "updated" date that matches the last real edit | 3 | Read the page |
Notes on the two checks people get wrong:
- A2. Training bots and answer bots are different. OpenAI's crawler documentation says opting out of OAI-SearchBot excludes you from ChatGPT search answers, while GPTBot covers model training. Anthropic separates ClaudeBot (training) from Claude-SearchBot (search indexing). Perplexity says PerplexityBot surfaces and links sites in results. Blocking training bots is a legitimate choice; blocking answer bots by accident is the most common way brands vanish. Give half points if robots.txt allows the bots but a firewall rule challenges them.
- A3. Google's own guidance lists nosnippet, data-nosnippet, max-snippet, and noindex as the controls for what appears in its AI features. A restrictive setting added years ago for another reason can quietly limit you today.
B. Does the page answer questions in liftable passages? (25 points)#
Answer engines quote passages, not whole pages. A page scores well here when any section could be lifted into an answer and still make sense.
| # | Check | Points | How to test |
|---|---|---|---|
| B1 | At least half the H2s are phrased as the questions buyers ask | 4 | Compare headings to your buyer question list |
| B2 | Each section opens with a 1-2 sentence direct answer | 5 | Read only the first two sentences under each H2 |
| B3 | Passages name entities in full instead of "it," "this," or "our tool" | 4 | Read a random paragraph in isolation |
| B4 | Comparisons use a real table with named products and concrete attributes | 4 | Award 0 if the page compares options in prose only |
| B5 | Time-sensitive claims are date-stamped ("as of October 2026") | 4 | Search the page for prices, versions, and "currently" |
| B6 | One clear topic per page, with a short summary near the top | 4 | Can you state the page's single question in one line? |
The fastest test for B2 and B3 together: copy the first two sentences of each section into a blank document. If someone who has not seen the page could understand each one, the page passes.
C. Does the page give engines evidence to quote? (20 points)#
Evidence is the highest-yield on-page fix. In the original GEO paper by Aggarwal et al. (KDD 2024), adding quotations raised visibility in generated answers by about 41%, adding statistics by about 33%, and citing sources by about 28%, while keyword stuffing scored about 9% below baseline.
| # | Check | Points | How to test |
|---|---|---|---|
| C1 | Key claims are backed by specific numbers with a named source | 5 | Highlight every claim; count how many have a number and a source |
| C2 | At least one attributed quote from a credible person or document | 4 | Quote with a name and role, or a quoted official document |
| C3 | External links go to primary sources (docs, papers, pricing pages) | 4 | Click each source link; award half if most go to secondary blogs |
| C4 | Original data or a first-hand test, with the method explained | 4 | Is there anything on the page a competitor could not copy? |
| C5 | No unverifiable or outdated statistics | 3 | Spot-check three numbers against their sources |
C5 is a deduction check in practice. A stale number that an engine repeats under your brand is worse than no number at all.
D. Do engines know exactly who you are? (10 points)#
Entity clarity means every source describes your brand the same way: same name, same category, same core facts. Engines combine sources, and inconsistent descriptions produce muddled or wrong answers.
| # | Check | Points | How to test |
|---|---|---|---|
| D1 | One consistent brand name and one-line category description across your site | 3 | Compare homepage, about page, footer, and pricing page |
| D2 | An about or company page with plain facts: what you do, for whom, founded, location, pricing model | 3 | Read it as a stranger |
| D3 | Organization and Product structured data that matches visible content | 2 | Google Rich Results Test; compare markup to page text |
| D4 | The same description on your LinkedIn, review profiles, app listings, and directories | 2 | Check your five most visible external profiles |
Keep D3 in proportion. Google states there is "no special schema.org structured data" needed for its AI features. Markup earns its 2 points by being accurate, not by existing.
E. Do trusted third parties mention you? (15 points)#
For evaluative questions ("best," "vs," "alternatives," "is it worth it"), engines draw heavily on what others say. Pew Research's March 2025 analysis found Wikipedia, YouTube, and Reddit were the most-linked sites in Google's AI summaries.
| # | Check | Points | How to test |
|---|---|---|---|
| E1 | You appear in the third-party comparison pages AI engines cite for your category | 5 | Run your top 10 buying questions; list cited URLs; check each for your brand |
| E2 | Review-site profiles are claimed, current, and accurate | 3 | Check the review sites cited in E1 |
| E3 | Your team participates in the Reddit threads AI cites, with disclosed accounts and within each subreddit's rules | 4 | List cited threads; check for a helpful, disclosed reply where allowed |
| E4 | Known inaccuracies about you in cited sources have been corrected or answered | 3 | Read what cited sources say about your pricing and features |
On E3: award zero for any undisclosed brand posting, purchased or multiple accounts, or vote manipulation, regardless of results. Reddit's rules require users to participate authentically and ban content manipulation. The guide to Reddit AEO explains how to do this well.
F. Are you measuring the outcome? (10 points)#
You cannot improve citations you do not measure, and AI answers vary too much to judge from one screenshot. A SparkToro and Gumshoe.ai study of 2,961 prompt runs found under a 1 in 100 chance of getting the same brand list twice from the same prompt.
| # | Check | Points | How to test |
|---|---|---|---|
| F1 | A fixed set of 30+ buyer questions, reused every cycle | 2 | Does the list exist and is it versioned? |
| F2 | Each question runs at least 3 times on each major engine | 2 | ChatGPT, Perplexity, Claude, Gemini, Google AI Overviews and AI Mode |
| F3 | Presence rate, citation rate, and share of voice tracked vs named competitors | 2 | See the AI share of voice formula |
| F4 | AI referral traffic is visible in analytics | 2 | GA4's AI Assistant channel or a custom channel group |
| F5 | Someone reads the answers for accuracy each cycle | 2 | Is there a log of wrong claims and fixes? |
Note on F4: GA4's AI Assistant channel covers sources like ChatGPT, Gemini, and Copilot and explicitly excludes Google's AI Overviews and AI Mode. Google's Search Generative AI performance reports in Search Console show impressions for those features. Use both. Our guide to AI visibility metrics explains what each number can and cannot tell you.
What does your GEO score mean?#
Add the six areas for each page. The score tells you what kind of problem you have, which tells you who should fix it.
| Score | Meaning | Fix first |
|---|---|---|
| 85-100 | Citation-ready | Keep measurement running; widen off-site presence |
| 70-84 | Solid, with gaps | The single lowest area, usually evidence or off-site |
| 50-69 | Readiness is holding you back | Structure (B) and evidence (C) |
| Under 50 | Something is blocking engines | Access (A), then everything else |
A worked example: scoring a comparison page#
Here is how the scoring plays out on a hypothetical "Tool A vs Tool B" page. The numbers are illustrative, not from a real audit.
| Area | Score | What we found |
|---|---|---|
| A. Access | 14/20 | PerplexityBot blocked by a firewall rule (A2 half points); comparison table renders only in JavaScript (A4 zero) |
| B. Structure | 18/25 | Good table, but sections open with marketing copy instead of answers (B2 zero); prices not date-stamped (B5 half) |
| C. Evidence | 9/20 | No numbers or source links behind pricing and performance claims (C1, C3 zero); one quote; a small first-hand test with no method (C4 half) |
| D. Entity | 8/10 | Review profiles describe an old product category (D4 zero) |
| E. Off-site | 6/15 | Absent from both comparison articles AI cites (E1 zero); no participation in cited threads (E3 zero) |
| F. Measurement | 6/10 | Question set exists, but runs only once per engine (F2 zero) and nobody logs accuracy (F5 zero) |
| Total | 61/100 |
The fix order follows the score: move the table into server-rendered HTML and fix the firewall rule (a developer task, one day), rewrite section openings and add sourced pricing (a writer task, one day), then start the slower off-site work. Those fixes recover A2, A4, B2, B5, and C1, which lifts the page from 61 to 79 on readiness alone. The measurement section then shows whether the higher score changes outcomes.
How do you run this audit faster?#
The access, structure, and evidence sections are quick: about 30-45 minutes per page once you are practiced. The measurement section is the slow part, because it requires dozens of questions times several engines times several runs, every month.
This is the part Sleeng automates. It generates buyer-intent questions from your URL, runs them on ChatGPT, Claude, Perplexity, Gemini, Google AI Overviews, and AI Mode, logs mentions and citations, calculates share of voice against your competitors, and finds the Reddit threads AI cites, which also covers most of checks E1 and E3.
For the strategy behind the checklist, read the pillar guide to answer engine optimization. For how the audit fits alongside classic SEO work, see GEO vs SEO.
Want your Section F scored automatically? Start a 14-day free trial of Sleeng and get a baseline of mentions, citations, and AI share of voice for your category across six AI surfaces.
Sources#
- GEO: Generative Engine Optimization, Aggarwal et al., arXiv 2311.09735 (KDD 2024)
- Google Search Central: AI features and your website
- Google Search Central Blog: Introducing Search Generative AI performance reports in Search Console (June 3, 2026)
- Google Analytics Help: Traffic-source dimensions and default channel group
- OpenAI: Overview of OpenAI crawlers
- Anthropic: Does Anthropic crawl data from the web, and how can site owners block the crawler?
- Perplexity: Perplexity crawlers
- Pew Research Center: Google users are less likely to click on links when an AI summary appears in the results (July 22, 2025)
- SparkToro: AIs are highly inconsistent when recommending brands or products (January 28, 2026)
- Reddit Rules
Frequently asked questions
How often should I run a GEO audit?
Run the full checklist quarterly on your top pages and whenever you redesign templates, change your CMS, or edit robots.txt. Those are the moments access and structure break. Run the measurement checks monthly, because AI answers in your category can change much faster than your pages do. New pages should pass the checklist before they are published.
Which pages should I audit first?
Audit the pages that answer questions buyers ask right before choosing: comparison pages, alternatives pages, pricing, product and category pages, and your best how-to guides. Five to ten pages is enough for a first pass. Blog posts that attract top-of-funnel traffic matter less, because evaluative AI answers tend to cite comparisons and discussions rather than general articles.
Do I need special tools to run a GEO audit?
No. The access checks need a terminal and Google Search Console. The structure and evidence checks need a careful read. The off-site and measurement checks need a spreadsheet and time to run your questions in each AI engine several times. Tools save time on the measurement section, which is the part that has to be repeated most often.
Is llms.txt part of a GEO checklist?
Not in this one. Google states that no AI text files or new machine-readable files are needed to appear in its AI features, and there is no public evidence that llms.txt changes which brands AI engines recommend. It can help coding agents read developer documentation. Publish one if it is cheap, but do not score pages on it.
What is a good GEO audit score?
In this checklist, 85-100 means a page is ready to be cited and the remaining work is off-site and ongoing. 70-84 is solid with clear gaps. 50-69 means structural or evidence problems are likely holding you back. Under 50 usually means an access or rendering problem, which should be fixed first because it blocks everything else.
Can a page score well and still not get cited?
Yes. The checklist measures readiness, not outcomes. Engines also weigh what other sites say about you, how many competing sources exist, and how each engine retrieves results, and answers vary between runs. That is why the checklist includes measurement checks: a high score with a low presence rate usually points to an off-site gap.
About the author
Founder, Sleeng
Dror founded Sleeng and COREPPC. He has spent 8+ years running large-scale PPC and performance programs for B2B, SaaS and ecommerce brands, and now studies how AI engines choose the brands and sources they cite.
Every statistic and price in this guide links to its original source and was checked on the date shown above.
