The AEO Audit: 12 Questions Every SEO Should Answer

First published: 19 June 2026 · Last updated: 19 June 2026

PILLAR 1

Technical access

1. Are AI crawlers explicitly allowed?
2. Is content rendered in initial HTML?
3. Is the site fast and lightweight for bot retrieval?

PILLAR 2

Content extractability

4. Are answers self-contained at section level?
5. Are questions explicitly stated as headings?
6. Are facts presented as discrete, citable units?

PILLAR 3

Entity & authority

7. Is the brand a recognised entity in Wikidata, Wikipedia, knowledge bases?
8. Are author bios and credentials machine-readable?
9. Is the brand mentioned by independent third-party sources?

PILLAR 4

Citation outcomes

10. Is the brand currently cited in ChatGPT, Perplexity, AI Overviews?
11. What is the citation share vs top 3 competitors?
12. Are citations growing or shrinking month-over-month?

0-9: Uncrawled

10-18: Visible

19-27: Citable

28-36: AI-native

The agency conversation about Answer Engine Optimisation has matured past "what is AEO" into "how do we audit for it". This article is the practitioner version of that audit. It complements the decision-maker overview on our sister site (the BMS post on the 7-step AEO audit for marketing teams) and goes deeper into the 12 questions an SEO needs to answer for any AEO engagement, whether that is a client diagnostic, an internal quarterly review, or a competitive intelligence pull. The audit is structured around four pillars because those are the four causal layers that determine whether a brand gets cited. Technical access is the prerequisite (if AI crawlers cannot reach the page, nothing else matters). Content extractability determines whether the page can be turned into a citable answer. Entity and authority signals determine whether the brand is trusted enough to be the citation. Citation outcomes are the lagging measurement. Optimising any of these layers in isolation produces uneven results. The 12-question audit forces sequential coverage. For broader context on the AEO content side, our existing AEO content framework and E-E-A-T 2026 guide cover what to write; this audit covers how to diagnose what is broken.

Pillar 1: Technical Access (Questions 1-3)

The technical access pillar is the gateway. AI engines cannot cite content they cannot fetch.

Question 1: Are AI crawlers explicitly allowed?

The audit step is to inspect `robots.txt` for explicit allow or disallow rules covering the major 2026 AI crawlers: GPTBot, OAI-SearchBot, ChatGPT-User (OpenAI), ClaudeBot (Anthropic), PerplexityBot, Perplexity-User (Perplexity), CCBot (Common Crawl), Google-Extended (Google AI training), Bytespider (ByteDance), and Bingbot (which feeds Microsoft Copilot). Scoring:

• 0 points: GPTBot, ClaudeBot, PerplexityBot are explicitly disallowed.
• 1 point: AI crawlers are not addressed at all (default behaviour, ambiguous).
• 2 points: AI crawlers are allowed via wildcard or absence of disallow.
• 3 points: AI crawlers are explicitly allowed with `User-agent: GPTBot \n Allow: /` style entries, demonstrating intent.

The most common failure mode: a 2023-era `robots.txt` with `User-agent: GPTBot \n Disallow: /` left in place from when the marketing team panicked about AI training. This single line is the most expensive AEO mistake on the audit. Our existing AI crawlers guide covers the bot landscape in depth.

Question 2: Is content rendered in initial HTML?

The audit step is to fetch the page with a vanilla user agent (curl, wget, or `view-source:`) and confirm the primary content is in the initial HTML response, not injected by client-side JavaScript. The reason: most AI crawlers in 2026 do not execute JavaScript. Content that requires JS to render is invisible to GPTBot, ClaudeBot, and PerplexityBot. Scoring:

• 0 points: Page is a JavaScript shell, primary content loads only after JS execution.
• 1 point: Some content is in HTML, but the answer-bearing sections (key paragraphs, lists, FAQ) are JS-injected.
• 2 points: All primary content is in initial HTML, but secondary content (related links, dynamic widgets) is JS.
• 3 points: Full page including all content is server-rendered or statically generated. AI crawlers see what users see.

Modern frameworks (Next.js with SSR or SSG, Astro, SvelteKit, traditional CMSs like WordPress) pass this question by default. Single-page React apps, Vue SPAs without SSR, and Webflow-style builders with heavy client-side hydration often fail. The fix is structural: migrate to SSR/SSG or pre-render the critical pages.

Question 3: Is the site fast and lightweight for bot retrieval?

AI crawlers operate on time budgets and prefer fast responses. A page that takes 8 seconds to TTFB will be deprioritised or skipped on revisits. This is the same dimension as Core Web Vitals but the relevant metric is server response time and total page weight, not LCP. Scoring:

• 0 points: TTFB over 2 seconds, total page over 5MB.
• 1 point: TTFB 1-2 seconds, page 2-5MB.
• 2 points: TTFB 500ms-1s, page 1-2MB.
• 3 points: TTFB under 500ms, page under 1MB, served from edge or CDN.

Cloudflare, Vercel Edge, and modern CDN setups make this trivial to fix. Legacy WordPress hosts on shared servers in the wrong region routinely fail this question.

Pillar 2: Content Extractability (Questions 4-6)

The content extractability pillar determines whether AI engines can turn the page into a discrete, citable answer.

Question 4: Are answers self-contained at section level?

AI engines cite at the passage level, not the page level. A passage that requires reading the previous three sections to make sense is not a useful citation candidate. The audit step is to take any major H2 section and read it cold. Does it stand alone? Does it answer a question without requiring prior context? Scoring:

• 0 points: Sections are narrative-dependent, references like "as we discussed above" or "as mentioned earlier" are common.
• 1 point: Some sections stand alone, but most assume reading order.
• 2 points: Most sections are self-contained with brief introductory framing.
• 3 points: Every section is engineered to be readable in isolation. Each H2 has a topic sentence that contains the answer; supporting paragraphs add evidence. The "lift any section out and it still makes sense" test passes.

The structural fix is to write every section as if it were a featured snippet candidate in itself: opening sentence answers the question, supporting sentences provide evidence, no anaphoric references to other sections.

Question 5: Are questions explicitly stated as headings?

AI engines match user prompts to content most efficiently when the prompt's question form matches a heading on the page. A page with H2s like "Pricing", "Features", "Customers" is harder to match than the same page with H2s like "How much does X cost?", "What features does X include?", "Who uses X?". Scoring:

• 0 points: All headings are noun-phrase labels, no questions.
• 1 point: A few questions in FAQ section only.
• 2 points: FAQ section uses questions, body H2s are mixed.
• 3 points: Body H2s and FAQ are both written as questions where the user query is in question form. Question form matches actual user prompts.

The discipline is to mine query data (Search Console queries, AlsoAsked, AnswerThePublic, real ChatGPT prompts) and convert the most common question forms into headings. This is the single highest-leverage extractability fix on most sites.

Question 6: Are facts presented as discrete, citable units?

AI engines prefer to cite specific facts (a number, a definition, a step in a process) over a paragraph of mixed assertion. A page full of long paragraphs requires the engine to extract a fact and re-state it in its own words, which reduces citation likelihood. A page that presents facts as discrete units (a single-sentence definition, a stat with a source, a numbered step, a comparison table cell) is easier to cite verbatim. Scoring:

• 0 points: Long paragraph blocks, facts buried in prose.
• 1 point: Some lists and tables, but most facts are still embedded in paragraphs.
• 2 points: Most factual content is structured (lists, tables, callouts) with prose providing context.
• 3 points: Every key fact has its own structural unit (a sentence, a table row, a callout box, a step). Citations show up as direct quotes.

Unextractable

"In our experience working with B2B SaaS clients across the SG market, we have generally found that the typical sales cycle from MQL to closed-won tends to range somewhere between three and six months for mid-market deals, although enterprise deals often take longer due to procurement processes and stakeholder alignment requirements."

Hard to cite. Fact is buried, qualified, and would need to be paraphrased.

Extractable

Average B2B SaaS sales cycle in SG (mid-market): 3-6 months from MQL to closed-won. Enterprise deals typically extend to 6-12 months due to procurement and stakeholder alignment. (Source: BestSEO 2026 SG B2B SaaS benchmark, n=23 clients)

Easy to cite. Fact is fronted, sourced, and can be lifted verbatim.

Pillar 3: Entity and Authority Signals (Questions 7-9)

The entity pillar determines whether the brand is recognised as a trusted source in the AI engines' underlying knowledge graphs.

Question 7: Is the brand a recognised entity in Wikidata, Wikipedia, knowledge bases?

AI engines lean heavily on structured knowledge bases (Wikidata, Wikipedia, Crunchbase, LinkedIn, industry-specific knowledge graphs) to validate that a brand is real, what it does, and what to attribute to it. A brand with no Wikidata entry and no Wikipedia article is invisible at the entity layer. Audit step: search Wikidata for the brand. Search Wikipedia. Search Crunchbase. Search the major industry directories relevant to the vertical. Scoring:

• 0 points: No entries in Wikidata, Wikipedia, or major directories.
• 1 point: Entry in one of the above (typically LinkedIn or Crunchbase).
• 2 points: Entry in two or three of the above, including Wikidata.
• 3 points: Wikipedia article (notable enough to survive deletion review), Wikidata entry with full property coverage, Crunchbase, LinkedIn, plus relevant industry directories.

For most SG SMEs, a Wikipedia article is unattainable (notability bar is high). Wikidata and Crunchbase are achievable for any business that publishes proof of existence. The fix is to systematically populate the structured knowledge layer with consistent NAP (name, address, phone) and brand metadata.

Question 8: Are author bios and credentials machine-readable?

AI engines weight content from authors with verifiable credentials more heavily than anonymous content. The audit step is to inspect the author bylines: do they link to a dedicated author page? Does the author page contain bio, credentials, links to their professional profiles? Is the author marked up with `Person` schema and `sameAs` references to LinkedIn, Twitter/X, Google Scholar? Scoring:

• 0 points: Anonymous content, no author bylines.
• 1 point: Author bylines but no author pages or credentials.
• 2 points: Author pages with bios but no schema markup.
• 3 points: Author pages with full bios, credentials, schema markup including `Person` with `sameAs` to verified profiles, and content history showing topical specialisation.

This is one of the highest-leverage fixes for content brands and the lowest-leverage for transactional brands. A SaaS company's product pages do not need elaborate author bios. A health publisher's articles absolutely do.

Question 9: Is the brand mentioned by independent third-party sources?

AI engines weight brand mentions across the web heavily for trust calibration. A brand mentioned by 47 independent sources (industry blogs, podcasts, news articles, conference recaps) is a more credible citation candidate than one mentioned only by its own properties. Audit step: search the brand name in Google, search in Bing, query ChatGPT for "what do people say about [brand]", check podcast appearances, conference speaker lists, industry award shortlists. Scoring:

• 0 points: Brand only appears on own properties and paid placements.
• 1 point: A handful of independent mentions (under 10).
• 2 points: Steady stream of independent mentions (20-50 per quarter).
• 3 points: High frequency of independent mentions including from authoritative sources (industry trade press, recognised podcasts, academic citations, conference programmes).

The fix is digital PR, podcast outreach, conference speaking, and contributing to industry discourse. This is the slowest pillar to fix but compounds the most.

Pillar 4: Citation Outcomes (Questions 10-12)

The citation outcomes pillar measures whether the AEO work is producing results.

Question 10: Is the brand currently cited in ChatGPT, Perplexity, AI Overviews?

The audit step is to run a tracked prompt set (20-50 prompts representative of customer queries) across ChatGPT (with web search), Perplexity, Google AI Overviews, and Microsoft Copilot. Record citation presence per prompt per engine. Scoring:

• 0 points: Zero citations across the prompt set on any engine.
• 1 point: Citations on under 20% of prompts on any single engine.
• 2 points: Citations on 20-40% of prompts across multiple engines.
• 3 points: Citations on over 40% of prompts across multiple engines, with the brand appearing in the first or second cited source position frequently.

This is the lagging metric. A brand that scores well on the first 9 questions but poorly here is usually within 60-90 days of seeing citation growth as the AEO work compounds.

Question 11: What is the citation share versus the top 3 competitors?

Run the same tracked prompt set with each of the top 3 competitors substituted into the recording. Calculate citation share: of prompts where any of you four are cited, what share goes to the brand under audit? Scoring:

• 0 points: Citation share under 10%.
• 1 point: Citation share 10-25%.
• 2 points: Citation share 25-50%.
• 3 points: Citation share over 50%, brand is the most-cited in the competitive set.

The competitive frame matters because AEO is zero-sum at the prompt level. If a competitor is consistently cited and you are not, the gap is informative: study what they have done structurally that you have not.

Question 12: Are citations growing or shrinking month-over-month?

The trend is more decisive than the absolute level. A brand with 12 citations growing 30% MoM is in a stronger position than a brand with 40 citations declining 5% MoM. Track the prompt set monthly and compute the trajectory. Scoring:

• 0 points: Citations declining MoM for 2+ months.
• 1 point: Citations flat MoM.
• 2 points: Citations growing 5-15% MoM.
• 3 points: Citations growing over 15% MoM consistently.

Trend data requires at least 90 days of measurement to be meaningful. New AEO programmes should expect 60-90 days of flat or near-zero citations before the first inflection.

Tier

Score

Profile

Priority next 90 days

Uncrawled

0-9

AI crawlers blocked or content invisible. Brand non-existent at entity layer.

Fix Pillar 1 (technical access). Nothing else matters until crawlers can read the site.

Visible

10-18

AI crawlers can read but content is hard to extract; weak entity signals.

Fix Pillar 2 (content structure) + Pillar 3 (entity baseline: Wikidata, author bios).

Citable

19-27

Citation candidates exist; lagging measurement starts to produce wins.

Scale Pillar 3 authority work + tighten Pillar 2 extractability per high-value query.

AI-native

28-36

Brand is consistently cited, leads competitive set, growing MoM.

Defend, expand prompt coverage, optimise per-engine (Perplexity vs ChatGPT vs AI Overviews).

Running the Audit: Process and Time-Boxing

The 12-question audit takes roughly 4-6 hours for a competent practitioner on a mid-sized site (under 500 indexed pages). For larger sites or competitive intelligence engagements, multiply by 1.5-2x. The breakdown:

1. Hour 1: Pillar 1 (technical). robots.txt review, view-source spot checks on 10 representative URLs, TTFB measurement via WebPageTest or DevTools.
2. Hour 2: Pillar 2 (extractability). Read 10 representative URLs cold. Score self-contained sections, question-form headings, fact-unit density.
3. Hour 3: Pillar 3 (entity + authority). Wikidata, Wikipedia, Crunchbase, LinkedIn searches. Author byline audit. Brand-mention scan (Google, Bing, ChatGPT recall query).
4. Hours 4-5: Pillar 4 (citations). Build the tracked prompt set (20-50 prompts). Run across ChatGPT, Perplexity, Google AI Overviews, Copilot. Record citations and competitor citations.
5. Hour 6: Scoring and remediation roadmap. Sum scores, identify the lowest-scoring questions, build a 90-day remediation plan prioritised by impact and effort.

The deliverable is a one-page scorecard plus a remediation roadmap with three priority tiers: fix this month, fix this quarter, fix over 90 days. Send to the client with an explicit pre/post measurement plan: re-audit at day 60 and day 90, expect detectable improvement on Pillar 4 by day 90 if Pillars 1 and 2 fixes ship in the first 30 days.

A Worked Example: SG B2B Manufacturing Site

Concrete worked example. Client: SG B2B manufacturing, $4,800/mo retainer, AEO audit on month two of engagement. Initial scores (out of 36):

• Pillar 1 (technical): 4/9 (GPTBot disallowed, content rendered server-side, slow TTFB)
• Pillar 2 (extractability): 3/9 (long paragraphs, label-style headings, prose-heavy)
• Pillar 3 (entity): 4/9 (LinkedIn yes, Wikidata no, no author bios on blog)
• Pillar 4 (citation): 1/9 (cited on 3 of 24 tracked prompts, 8% citation share, no trend data yet)
• Total: 12/36 — Visible tier, lower bound.

90-day remediation roadmap:

• Month 1: Remove GPTBot/ClaudeBot disallows. Move from shared host to Cloudflare-fronted setup. Rewrite top 8 pillar pages with question-form H2s and self-contained sections. Add author bios with Person schema across blog.
• Month 2: Submit to Wikidata (created entity, full property coverage). Submit to 6 SG industry directories. Convert top 12 product pages to fact-unit format with comparison tables.
• Month 3: Begin digital PR programme targeting trade press. Re-run audit.

Day 90 re-audit scores:

• Pillar 1: 8/9 (technical fully fixed except CDN regional optimisation pending)
• Pillar 2: 7/9 (top 20 pages remediated, long-tail still pending)
• Pillar 3: 6/9 (Wikidata done, directories done, author bios done; PR is slower compounding work)
• Pillar 4: 4/9 (citations on 9 of 24 tracked prompts, 22% citation share, MoM trajectory positive)
• Total: 25/36 — Citable tier.

Six-month outlook: Pillar 4 expected to reach 6-7/9 as PR work compounds and the prompt-test set grows. Total score target by end of year: 30+/36 — AI-native tier.

Frequently Asked Questions

How is this different from a traditional SEO audit?

A traditional SEO audit checks crawlability for Googlebot, on-page optimisation for keyword targets, backlink profile, and Core Web Vitals. The AEO audit overlaps on technical crawlability (Pillar 1) but adds AI-specific bots, replaces keyword targeting with question-form headings and extractable facts (Pillar 2), adds entity-graph signals beyond domain authority (Pillar 3), and replaces ranking measurement with citation measurement across AI engines (Pillar 4). Many sites that score well on a traditional SEO audit fail Pillars 2 and 4 of the AEO audit because the optimisation goals are categorically different. We recommend running both audits separately for any client where AI search traffic matters.

How often should an AEO audit be re-run?

For active retainer clients, full re-audit at day 60 and day 90 of the initial engagement, then quarterly. The Pillar 4 prompt-test should be re-run monthly once the engagement is in steady state, with the full 12-question audit refreshed quarterly. The prompt set itself should be refreshed every quarter to reflect emerging customer queries (new product features, new vertical entries, market shifts).

Can I run the audit on a competitor?

Yes, and competitive AEO audits are one of the highest-leverage uses of the framework. Substitute the competitor as the audited brand and run all 12 questions. The Pillar 4 measurement (citation share) directly surfaces which competitor is winning AI visibility and on which queries. The Pillar 1-3 measurements identify what they have done structurally that you have not, which becomes the input to your remediation roadmap.

What tools do I need to run the audit?

Minimum stack: a browser with view-source, curl or wget, a robots.txt parser (or grep), Search Console for the audited site, a spreadsheet for the prompt-test recording. For Pillar 4 at scale, a paid AI visibility tool (Profound, AthenaHQ, Otterly) automates the prompt-test recording across engines. For Pillar 3, Wikidata and Wikipedia interfaces are free; Crunchbase has a free read tier. Total tooling cost ranges from zero (DIY across 4-6 hours) to $200-500/mo (with paid AI visibility tooling).

How does Pillar 3 (entity + authority) interact with E-E-A-T?

Heavy overlap. E-E-A-T (Experience, Expertise, Authoritativeness, Trust) is Google's framing of the same underlying signals that AI engines use for citation trust calibration. A brand with strong E-E-A-T scores well on Pillar 3 by definition. The AEO frame is more granular about which signals are machine-readable: a strong author with no Person schema and no `sameAs` links is opaque to AI engines even if their credentials are real. See our existing E-E-A-T 2026 guide for the broader treatment.

What is the relationship between this audit and a GEO audit?

AEO and GEO overlap heavily but are not identical. AEO focuses on getting cited by AI engines that produce direct answers (ChatGPT, Perplexity, Google AI Overviews, Copilot). GEO is broader and includes generative search experiences as a category, including image and video AI, vertical AI assistants, and emerging interfaces. The 12-question AEO audit covers the answer-engine subset; GEO-specific audits add questions on multimodal content, vertical assistant coverage (Shopify Magic, Notion AI, etc.), and emerging surface areas. See our GEO playbook for the broader GEO treatment.

Related reading

• AEO content framework: how to write content that scores Pillar 2
• E-E-A-T in 2026: the entity and authority foundation for Pillar 3
• AI SEO audit workflow: the broader AI-era audit methodology