AEO checklist

An AEO checklist is a list of checks that a website must pass before AI answer engines can find it, cite it, and describe it correctly. This one has 50 checks in ten groups, from crawler access and facts in the HTML to third-party sources and measurement. Each check says how to verify it, what counts as a pass, how strong the evidence is, and which source or guide explains it.

Answer engine optimization (AEO) is the work of making a company's pages and facts easy for AI answer engines to find, cite, and recommend. This checklist is the audit companion to the complete guide to answer engine optimization, which explains why each area matters; the mistakes to avoid are described in answer engine optimization antipatterns. The sources behind each check were read on September 27, 2026.

Scope

• Covers: unpaid visibility in AI answers from Google's AI Overviews and AI Mode, ChatGPT search, Gemini, Claude, Perplexity, and Microsoft Copilot.
• Does not cover: advertising inside AI answers, or technical SEO beyond what AI answers depend on. The planned technical SEO checklist for AI search covers the rest of technical SEO.
• For: marketers, founders, and developers at B2B companies. Most checks need access to the website's code or content system.
• Pass criteria are AEO HQ's recommendations. The facts behind them link to their sources.
• Evidence labels:
  • Strong: official platform documentation, or several independent studies that agree.
  • Moderate: laboratory studies, a single study, surveys, or correlations.
  • Weak: one small test, or a source with a conflict of interest.

Prerequisites

• Access to the site's robots.txt file, its host or CDN firewall settings, and its server logs.
• Verified properties in Google Search Console and Bing Webmaster Tools.
• An analytics tool, such as Google Analytics 4 (GA4).
• A list of 20 to 50 questions that buyers ask, in their own words. The AEO guide describes how to collect them.
• A one-page fact sheet: the company's name, a one-line description, prices, scope, deliverables, turnaround times, and the people who do the work.
• A way to fetch a page without running JavaScript, such as curl or the browser's "view source".

Quick checks to run first

If time is short, AEO HQ recommends starting with five checks. Each takes minutes, and a failure on any of them can keep pages out of AI answers whatever else is done:

1. Check 1.6: the search crawlers are allowed in robots.txt.
2. Check 1.8: the host and firewall let those crawlers through.
3. Check 1.2: Search Console includes the site in generative AI features.
4. Check 2.1: key facts are in the HTML the server sends.
5. Check 2.3: no noindex or nosnippet rule blocks the pages you want cited.

1. Crawling and indexing

These checks cover whether answer engines can reach and index the site. How ChatGPT, Gemini, Claude, Perplexity, and Copilot find and cite sources lists the index and crawlers behind each assistant.

• 1.1 Core pages are indexed in Google. Verify: run URL Inspection in Search Console on each core page. Pass: every core page is indexed. Why: to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet (opens in a new tab). Evidence: strong (official documentation).
• 1.2 Search Console includes the site in generative AI features. Verify: open Settings > Search generative AI in Search Console. Pass: the control is set to include, either directly or inherited from a parent property. Why: a site must be included in this control to be eligible for display in generative AI features on Google Search (opens in a new tab); inclusion is the default for all properties, and the control reached all websites worldwide on August 31, 2026 (opens in a new tab). Evidence: strong (official documentation).
• 1.3 Core pages are indexed in Bing. Verify: run URL Inspection in Bing Webmaster Tools. Pass: every core page is indexed. Why: Microsoft Copilot generates a search query and sends it to the Bing search service (opens in a new tab), and ChatGPT search sometimes partners with other search providers, and OpenAI's help page lists Microsoft's privacy statement among those providers' policies (opens in a new tab). Evidence: strong (official documentation).
• 1.4 A sitemap is submitted to both, with true dates. Verify: check the Sitemaps reports in Search Console and Bing Webmaster Tools, and compare each lastmod value with the page's last real edit. Pass: the sitemap is accepted, and lastmod changes only when the content changes. Why: Google uses lastmod only if it is "consistently and verifiably" accurate (opens in a new tab), and Bing says accurate lastmod values help it detect content changes (opens in a new tab). Evidence: strong (official documentation).
• 1.5 IndexNow is notified when pages are added, changed, or removed. IndexNow is a protocol that tells participating search engines about changed URLs. Verify: publish a test change and check the submission log. Pass: every publish, update, and removal sends a submission. Why: Bing asks site owners to use IndexNow when URLs are added, content is updated, or URLs are removed (opens in a new tab); IndexNow's FAQ lists endpoints for Amazon, Bing, Naver, Seznam.cz, Yandex, and Yep, but not Google, and says "Submitting a URL does not guarantee immediate indexing" (opens in a new tab). Evidence: strong for Bing's request; no study has measured the effect on AI answers.
• 1.6 robots.txt allows the search and user-triggered crawlers. Verify: read /robots.txt. For each of Googlebot, Bingbot, OAI-SearchBot, ChatGPT-User, Claude-SearchBot, Claude-User, and PerplexityBot, find the group that names it, or the * group if none does. Pass: no Disallow rule covers a page you want cited. Why: sites that opt out of OAI-SearchBot will not be shown in ChatGPT search answers, and robots.txt changes can take about 24 hours to take effect (opens in a new tab); disabling Claude-SearchBot may reduce a site's visibility and accuracy in Claude's search results (opens in a new tab); Perplexity recommends allowing PerplexityBot to appear in its search results (opens in a new tab); and a crawler obeys the group that matches it, using the * group only when no group matches (opens in a new tab). Evidence: strong (official documentation and the Internet standard).
• 1.7 Training crawlers and Google-Extended are set on purpose. Verify: read the robots.txt rules for GPTBot, ClaudeBot, and Google-Extended. Pass: each rule matches a written decision. Why: GPTBot crawls content that may be used to train OpenAI's models, and each OpenAI crawler setting is independent of the others (opens in a new tab); ClaudeBot collects content that could contribute to training Anthropic's models (opens in a new tab); Google-Extended controls Gemini training and grounding in Gemini Apps, and does not affect inclusion in Google Search (opens in a new tab). Evidence: strong for what each control does; no study has measured whether allowing training crawlers changes AI recommendations.
• 1.8 The host, CDN, and firewall let crawlers through. Verify: review the bot rules in your CDN or firewall, and confirm in server logs that the crawlers in check 1.6 receive HTTP 200 responses. Pass: no blocks or challenges for verified crawler IP addresses. Why: OpenAI asks site owners to confirm that the host or CDN allows traffic from its published searchbot IP addresses (opens in a new tab); Google lists allowing crawling "in robots.txt, and by any CDN or hosting infrastructure" among the basics for its AI features (opens in a new tab); on Vercel, the AI Bots ruleset can deny all traffic identified as AI bots, and the Bot Protection ruleset can challenge unwanted bots; both are inactive by default (opens in a new tab). Evidence: strong (official documentation).
• 1.9 Every important page has a crawlable internal link. Verify: crawl the site from the home page and compare the result with the sitemap. Pass: no core page is missing from the crawl. Why: Google discovers new pages when it extracts a link from a known page, and from sitemaps (opens in a new tab); Bing lists crawlable internal links among the ways it discovers URLs (opens in a new tab). Evidence: strong (official documentation).
• 1.10 Key pages appear in Brave Search. Verify: search Brave for the company name and for the titles of key pages. Pass: they appear. Why: Anthropic lists Brave Search as a web search subprocessor (opens in a new tab), and Brave's crawler does not crawl pages that Googlebot cannot crawl (opens in a new tab). Evidence: strong for the documentation; Anthropic does not document how Claude uses Brave's results. See how to rank in Claude.

2. Facts in the HTML

• 2.1 Key facts are in the HTML the server sends. Verify: fetch each key page without running JavaScript. Pass: prices, scope, deliverables, turnaround, the author, and the dates all appear in the raw HTML. Why: in Vercel's network data from December 2024, none of the major AI crawlers rendered JavaScript (opens in a new tab), and Bing warns against hiding critical content behind client-side rendering, saying content that cannot be reliably rendered may not be indexed or selected for grounding (opens in a new tab). Evidence: strong.
• 2.2 Key answers are in the visible page text. Verify: read each key page with tabs and menus closed. Pass: no key answer exists only inside a tab, an expandable menu, a PDF, or an image. Why: Microsoft advises against hiding important answers in tabs or expandable menus, relying on PDFs for core information, or putting key information only in images (opens in a new tab). Evidence: moderate (platform guidance in a blog post).
• 2.3 No indexing or snippet rule blocks pages you want cited. Verify: check each page's robots meta tag and X-Robots-Tag header. Pass: no noindex, nosnippet, max-snippet:0, NOARCHIVE, or NOCACHE on those pages. Why: Google names nosnippet, data-nosnippet, max-snippet, and noindex as the controls that limit what its AI features show (opens in a new tab); Bing says NOARCHIVE keeps content out of Copilot responses and NOCACHE limits Copilot to the URL, title, and snippet (opens in a new tab). Evidence: strong (official documentation).
• 2.4 Facts in structured data also appear as visible text. Verify: compare each value in the page's JSON-LD with the visible page. Pass: every value in the markup is also in the text. Why: in an October 2025 test, AI chatbots did not use JSON-LD markup when fetching a page directly and extracted only visible HTML (opens in a new tab) (practitioner test), and Google asks that structured data match the visible text on the page (opens in a new tab). Evidence: weak for the test; strong for Google's rule.

3. Page content

How to write these pages is covered in how to write content for answer engines.

• 3.1 Each priority buyer question has one page. Verify: map every question on your list to one URL. Pass: each question or intent has one page, and no page is a near-copy of another. Why: Bing says URLs focused on one primary topic are more likely to be selected for grounding results (opens in a new tab); Google says creating separate content for every possible variation of how people search, primarily to manipulate rankings or generative AI responses, violates its scaled content abuse policy (opens in a new tab). Evidence: strong (official documentation).
• 3.2 The title, H1, URL, and first paragraph use the buyer's wording. Verify: compare each with the question it answers. Pass: the main terms of the question appear in all four. Why: Bing asks site owners to align titles, headings, and content intent (opens in a new tab), and in controlled experiments, language models relied heavily on how relevant a website was to the query while largely ignoring features such as scientific references or a neutral tone (opens in a new tab) (peer-reviewed). Evidence: moderate.
• 3.3 The answer comes first. Verify: read the first three sentences of each page. Pass: they answer the question without the rest of the page. Why: Bing asks site owners to place essential information near the top and to avoid long introductions (opens in a new tab). The one controlled field study of answer-first changes found a 1.82-fold rise in ChatGPT referrals, but a placebo test gave p = 0.16, so the effect is "suggestive, not conclusive" (opens in a new tab) (preprint; one site; the authors work for the site's owner). Evidence: strong for Bing's guidance (official documentation); weak for the measured effect.
• 3.4 The page states specific facts. Verify: look for prices, scope, specifications, and comparisons written as numbers and names. Pass: a buyer can learn what it costs, what is included, and how long it takes without contacting sales. Why: in 252,000 controlled trials across six models, explicit price information and a recent timestamp consistently raised the chance of being cited first (opens in a new tab) (peer-reviewed; laboratory; the authors work for a software company), and 27% of U.S. B2B professionals who use AI at work said AI vendor recommendations don't reflect real pricing or contract structures (opens in a new tab) (vendor survey; 519 respondents). Evidence: moderate.
• 3.5 Each key statement stands on its own. Verify: read key sentences out of context. Pass: facts and definitions are explicit, and no key statement depends on an earlier paragraph or an image. Why: Bing says pages are more likely to be selected for grounding queries and citations when content stands on its own, with facts and definitions explicit (opens in a new tab). Evidence: strong (official documentation).
• 3.6 People write and review every page. Verify: check the publishing record for a named author and a reviewer. Pass: no page was mass-generated or rewritten in bulk by a language model without review. Why: Bing says large-scale content generated without oversight or editorial review may be excluded from indexing (opens in a new tab), and in a full generative search pipeline, existing GEO approaches often degraded visibility in retrieval and reranking (opens in a new tab) (peer-reviewed). Evidence: strong.
• 3.7 No keyword stuffing or engineered language. Verify: read key pages aloud and look for repeated phrases. Pass: terms appear only where they read naturally. Why: Bing lists keyword stuffing and "artificially engineered language" written to trigger citations or AI responses among practices that can reduce visibility or lead to removal (opens in a new tab). Evidence: strong (official documentation).

4. Dates and authors

• 4.1 Each page shows when it was published and last updated. Verify: view the page. Pass: both dates are visible. Why: a recent timestamp consistently raised the chance of being cited first in controlled trials (opens in a new tab) (laboratory), and Bing asks site owners to update content when facts or guidance change (opens in a new tab). Evidence: moderate.
• 4.2 Dates change only when the content changes. Verify: compare the visible and lastmod dates with the edit history. Pass: no date moved without a substantive edit. Why: Google says lastmod should reflect the last significant update to a page (opens in a new tab). Evidence: strong for sitemaps; no study has measured the effect of false dates on AI answers.
• 4.3 Each author has a profile page, marked up as a person. Verify: follow the byline link, and check the page's Article markup. Pass: the profile says who the author is, and the markup gives the author as a Person with a url or sameAs link. Why: Google asks for an author URL that uniquely identifies the author, such as a bio page, and accepts sameAs as an alternative (opens in a new tab). Evidence: strong for Google's documentation; weak for any effect on AI answers.

5. Entity consistency

An entity is a distinct thing, such as a company or a person, that search systems try to tell apart from others. The planned entity and brand consistency checklist covers this group in more depth.

• 5.1 The company uses one name and one description everywhere. Verify: compare the site, its structured data, social profiles, directories, and review sites with the fact sheet. Pass: the name, description, and core facts match on each. If another company or person shares the name, pages state a distinguishing detail near it, such as the domain or the category. Why: Bing asks site owners to use clear and consistent naming for people, organizations, products, and locations (opens in a new tab), and language models "often yield ambiguous answers or incorrectly merge information belonging to different entities" that share a name (opens in a new tab) (peer-reviewed). Evidence: moderate.
• 5.2 The home page has Organization markup that links the official profiles. Verify: test the home page with Google's Rich Results Test or read its source. Pass: the markup gives the name, logo, and URL, with sameAs links to the official profiles. Why: Google says Organization markup can help it disambiguate an organization in search results, and that sameAs can point to its profile pages on social media or review sites (opens in a new tab). Evidence: strong for Google's documentation; weak for any effect on AI answers.

6. Structured data

Structured data, also called schema markup, is code that labels a page's facts for software.

• 6.1 Markup matches the visible text. Verify: compare the markup with the page. Pass: no value in the markup is missing from, or different from, the page. Why: Google asks that structured data match the visible text (opens in a new tab), and Bing says markup must accurately reflect visible content (opens in a new tab). Evidence: strong (official documentation).
• 6.2 Markup is treated as upkeep, not as a way to earn citations. Verify: review the plan and any vendor proposals. Pass: no item promises citations from markup alone. Why: Google says structured data isn't required for generative AI search and there is no special schema.org markup to add (opens in a new tab); after 1,885 pages added JSON-LD, their AI Overview citations fell 4.6% against matched controls, and changes in AI Mode and ChatGPT were indistinguishable from zero (opens in a new tab) (vendor study). Evidence: moderate.
• 6.3 FAQ markup is optional. Verify: check whether the plan relies on FAQPage markup. Pass: visible FAQs stay for readers, and no result depends on FAQ markup. Why: Google's FAQ rich result stopped appearing in Google Search starting May 7, 2026 (opens in a new tab). Evidence: strong (official documentation).
• 6.4 llms.txt is optional. Verify: if the site has an llms.txt file, compare it with the site. Pass: it matches the site, and no plan depends on it. Why: 97% of llms.txt files received no requests in May 2026 (opens in a new tab) (vendor study; server logs from 137,210 domains), and Google says these files aren't needed for Google Search and won't negatively or positively affect visibility or rankings (opens in a new tab). Evidence: moderate.

7. Third-party sources

What other sites say about a company is covered in brand mentions and AI recommendations.

• 7.1 The company has accurate profiles on the review sites and directories in its category. Verify: list the third-party sites that appear in AI answers to your buyer questions, and check the company's profile on each. Pass: each profile exists, is complete, and matches the fact sheet. Why: 45% of B2B software buyers said citations from software review sites are the most confidence-inspiring signal in an AI answer (opens in a new tab) (vendor survey; 1,076 respondents, March 2026). Evidence: moderate.
• 7.2 Reviews are real and are not paid for by sentiment. Verify: review how the company asks for reviews. Pass: no incentive depends on a positive review, and staff reviews disclose the connection. Why: the U.S. Federal Trade Commission's rule prohibits fake reviews, including AI-generated ones, and incentives conditioned on reviews expressing a particular sentiment (opens in a new tab). Evidence: strong (regulation).
• 7.3 Independent sites describe the company. Verify: search for the company name on sites it does not control. Pass: editorial articles, comparisons, or industry lists describe the company accurately. Why: in API tests, 86% to 95% of the domains that ChatGPT and Claude cited were earned media, such as reviews and editorial coverage (opens in a new tab) (preprint), and across 75,000 brands, branded web mentions correlated with AI visibility at 0.656 to 0.709, while link metrics showed "very weak correlations" (opens in a new tab) (vendor study; correlation, not cause). Evidence: moderate.
• 7.4 Staff take part in communities openly. Verify: review staff posts on forums such as Reddit. Pass: every post by staff discloses the connection to the company. Why: for newly launched startups, visibility on Perplexity correlated with community presence on Reddit (r = +0.395) (opens in a new tab) (preprint), and Google says seeking inauthentic mentions across the web "isn't as helpful as it might seem" (opens in a new tab). Evidence: weak to moderate.
• 7.5 No bought placements, and no company-run site posing as independent. Verify: review marketing spend and the company's own review or comparison pages. Pass: there are none. Why: the FTC rule prohibits a business from misrepresenting that a website it controls provides independent reviews of a category that includes its own products (opens in a new tab), and Bing treats link buying and "artificial social promotion schemes that simulate popularity" as violations (opens in a new tab). Evidence: strong (regulation and official documentation).

8. Offer pages

• 8.1 The offer page states price, scope, deliverables, and turnaround as text. Verify: fetch the page's raw HTML. Pass: all four appear. Why: Perplexity says its answer engine is "more likely to recommend merchants who provide deeper product details such as availability, reviews, pricing, and specifications" (opens in a new tab) (official documentation, for shopping). Evidence: moderate for B2B services.
• 8.2 A buyer or an AI agent can reach checkout without a login wall or bot challenge. Verify: go through the purchase path in a clean browser, and review bot-protection rules on those pages. Pass: no CAPTCHA, required login, or firewall block before payment. Why: Stripe asks AI agents to report what stopped a purchase, with tags that include captcha, anti_bot_script, waf_block, and login_required (opens in a new tab). Evidence: strong for the documentation; the effect on AI recommendations has not been measured. See how AI agents find and buy services.

9. Measurement

AEO HQ's measurement design describes these checks in detail.

• 9.1 A fixed set of buyer questions is run on a schedule, several times each. Verify: check the measurement log. Pass: the same questions run on each engine at least weekly, with several runs per question. Why: asked the same question 100 times, ChatGPT or Google's AI has less than a 1 in 100 chance of giving the same list of brands in any two responses (opens in a new tab) (industry study; 2,961 runs), and the standard error of a brand's detection rate fell from 0.370 with one run per prompt to 0.081 with seven (opens in a new tab) (preprint; the lead author is affiliated with a tracking vendor). Evidence: strong. See how to measure AI visibility.
• 9.2 Results are reported as rates with intervals, for each engine. Verify: read the report. Pass: each engine shows the share of runs that name the company, with an interval suited to small samples. Why: with fewer than a few hundred data points, methods based on the central limit theorem usually produce error bars that are too small (opens in a new tab) (peer-reviewed). AEO HQ recommends Wilson score or Bayesian intervals, as described in our methodology. Evidence: strong.
• 9.3 Tests use the consumer apps, not only APIs. Verify: check that each run records the product, mode, and account state. Pass: at least a monthly sample runs in the consumer apps. Why: for the same queries, the API and the consumer interface shared only 12.0% (ChatGPT) and 14.8% (Gemini) of cited domains (opens in a new tab) (preprint). Evidence: moderate.
• 9.4 Answers about the company are checked for accuracy. Verify: grade answers to branded questions against the fact sheet. Pass: errors are logged and traced to a source. Why: 11.0% of 98,020 claims in Google AI Overviews were not supported by the pages they cited (opens in a new tab) (preprint). Evidence: moderate.
• 9.5 Search Console's generative AI performance report is reviewed. Verify: open the report. Pass: impressions are recorded each month. Why: the report shows impressions of a site's links in Search generative AI features, and reached all websites worldwide on August 31, 2026 (opens in a new tab). Evidence: strong (official documentation).
• 9.6 Bing Webmaster Tools' AI Performance report is reviewed. Verify: open the report. Pass: citations, cited pages, and grounding queries are recorded each month. Why: the report shows citation counts, cited pages, and grounding queries, which are the key phrases the AI used when it retrieved the cited content (opens in a new tab). Grounding means basing an answer on retrieved sources. Evidence: strong (official documentation).
• 9.7 Analytics count AI referrals in a channel of their own. Verify: check the channel settings in GA4 or your analytics tool. Pass: a channel covers chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, and copilot.microsoft.com. Why: GA4's default AI Assistant channel covers sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok, and excludes AI Overviews and AI Mode, which count as Organic Search (opens in a new tab); ChatGPT adds utm_source=chatgpt.com to referral URLs (opens in a new tab). Evidence: strong. See how to track AI referral traffic in GA4.
• 9.8 Server logs record visits from AI crawlers. Verify: filter logs by the user agents in check 1.6, and match the IP addresses against the published lists. Pass: monthly counts for each crawler and URL. Why: Anthropic publishes the IP addresses its crawlers use (opens in a new tab), and OpenAI publishes an IP address list for each of its crawlers (opens in a new tab). Evidence: strong (official documentation).
• 9.9 Buyers are asked where they heard of the company. Verify: check the signup, contact, or checkout form. Pass: it has a free-text "How did you hear about us?" field. Why: in one agency's records, first-touch attribution credited AI with only 28 of the 189 leads that named an AI tool in that field (opens in a new tab) (a single company). Evidence: weak.

10. Practices to avoid

Each item here passes when the practice is absent. The full list, with a test for each, is in answer engine optimization antipatterns and generative engine optimization antipatterns.

• 10.1 No hidden text or instructions aimed at AI. Verify: search the HTML, alt text, and structured data for text that visitors cannot see. Pass: none. Why: Google's spam policies cover attempts to manipulate generative AI responses (opens in a new tab) and list hidden text, such as white text on a white background; Bing says attempts to manipulate language models used by Bing or Copilot may result in reduced visibility or removal (opens in a new tab). Evidence: strong (official documentation).
• 10.2 No pages made for every variation of a query, city, or industry. Verify: list pages built from one template. Pass: each page answers a distinct question (check 3.1). Why: Google treats creating separate content for every variation, primarily to manipulate rankings or AI responses, as scaled content abuse (opens in a new tab). Evidence: strong (official documentation).
• 10.3 No faked signals. Verify: checks 4.2, 7.2, and 7.5. Pass: no false dates, fake reviews, or bought mentions. Why: see those checks. Evidence: strong.
• 10.4 No promises of guaranteed placement. Verify: read the company's own claims and those of any vendor it hires. Pass: no one promises citations, rankings, or a "#1" position in an AI answer. Why: Google says third-party tools "can't guarantee performance" and lists "AEO" and "GEO" tools among services to be wary of (opens in a new tab); OpenAI says of ChatGPT search results, "Placement is not guaranteed" (opens in a new tab); Bing says "GEO does not guarantee grounding or citations in AI experiences" (opens in a new tab). Evidence: strong (official documentation).

Version and change log

• Version 1.0, September 28, 2026: First published. Sources checked on September 27, 2026.

Next steps

AEO HQ's Instant AEO Audit ($499) automates a small part of this checklist: it reads robots.txt rules for seven crawler names, checks for llms.txt and a sitemap, and records the home page's structured data types and robots meta tag. Its limits are listed in our methodology. AEO HQ also sells fixed-price services with published prices: a $2,500 audit, a $4,995 blueprint, and an $8,995 package that adds technical implementation to the blueprint. See what each AEO service includes.