AI Search Visibility

AEO Audit: The 25-Point Checklist for B2B SaaS Sites

Stefka Team October 12, 2026 8 min read

An AEO audit checks whether AI answer engines can reach your content, understand it, trust it and cite it. It is a diagnostic, not a fix: the output is a ranked list of what is currently stopping ChatGPT, Google's AI Overviews, Perplexity and Claude from naming you, and which of those things is worth your time.

The reason to write this down is that almost nobody has. We pulled the full Google results for aeo audit on 12 October 2026, twenty positions deep. Fourteen of the eighteen organic results sell a tool or a free audit. Three explain the method. Google's own AI Overview sits above all of them and cites ten sources, six of which are audit products citing their own marketing pages.

What does an AEO audit actually check?

Twenty-five checks, in four groups. The order matters: a failure in group one makes groups two to four irrelevant, because nothing downstream can help a page an AI crawler cannot fetch.

Group 1 — Access (5 checks)

1. robots.txt does not block GPTBot, ClaudeBot, PerplexityBot, Google-Extended or CCBot. Blocking these is the single most common silent failure, usually inherited from a template or a well-meaning security review.
2. The page returns 200 to a crawler without JavaScript. Parse your own pages with JS disabled. If the content appears only after hydration, assume it is not being read.
3. No login, consent wall or geo-gate sits in front of the content.
4. The page is in the XML sitemap, and the sitemap is declared in robots.txt.
5. llms.txt, if present, is accurate rather than aspirational. It is not a ranking factor and no engine has committed to honouring it; an inaccurate one is worse than none.

Group 2 — Structure (7 checks)

6. The first 40 words answer the page's question directly, in a form that stands alone when lifted out.
7. Every heading below the lede is a question or a specific claim, and the first sentence under it answers that heading.
8. Each section is self-contained — it makes sense quoted without the paragraph before it.
9. Facts appear as subject-predicate-object statements, not as implications across three sentences.
10. Numbers are in text, never in images.
11. Comparison data is in real HTML tables with text cells.
12. There is a visible FAQ of four to six entries with answers of roughly 40 to 60 words.

Group 3 — Trust and entity clarity (7 checks)

13. Organization and Person schema exist and are consistent with what the rest of the web says about you.
14. sameAs points at the profiles you actually control.
15. The brand name is disambiguated. This is not theoretical. Our own domain has been co-cited with a baby-name dictionary, an ancestry site and a SoundCloud profile for ten consecutive weekly measurements, because "Stefka" is a given name before it is a studio.
16. Every statistic names its source and its date, with a link.
17. Claims about clients are attributable. An unverifiable case study is a liability in a system that weighs corroboration.
18. Author and organisation are identifiable — a named person with a traceable history.
19. Third-party corroboration exists. What other sites say about you is the part you cannot write yourself, and the evidence suggests it matters more than anything on this list.

Group 4 — Measurement (6 checks)

20. You know what AI engines currently say about you, across a prompt set that matches real buyer questions rather than the ones you would like to win.
21. You know what they say about your competitors on the same prompts.
22. You know which sources those answers are built from. This is the lever. If an answer is assembled from three domains, being on those three domains matters more than anything on your own site.
23. The prompt set is fixed so that movement is measurable rather than an artefact of changing the questions.
24. You re-measure on a schedule, not when someone asks.
25. You separate ranking from citation. They are different systems. We have measured a firm with 3,797 monthly US organic visitors being named by Google's AI while a competitor ranking at organic position 2 was not.

The technical layer in detail covers groups one and two check by check, including what changed in 2026.

Is a GEO audit the same thing?

Effectively, yes, and the labels are not worth managing.

Two things are worth taking from that table.

The first is that the vocabulary barely exists. Fourteen phrasings a practitioner would reach for return no measured search volume at all. Three have any. If you are writing about this, you are writing for a category that has not finished naming itself.

The second is that all three real terms turned upward together. Between August and September 2026 the assistant volumes went 3 → 6, 3 → 10 and 3 → 11 — three independent terms moving in the same month. On Google, aeo audit ran at 40 to 50 searches a month through late 2025 and reached 210 in August 2026, with no month down since January. Something changed in how often people ask this, and it changed recently.

We use "AEO" in titles and URLs because Google volume is growing fastest on that phrasing, and treat GEO and LLM SEO as synonyms in the copy. What answer engine optimization is sets out the wider vocabulary question if you want it.

Why does the search result for this look so strange?

Because Google has no entity model for the phrase.

The People Also Ask block on aeo audit carries 29 questions. Two of them are about AEO audits. The rest walk away from the subject within two hops: first into the careers of AI auditors — "How do I become an AI auditor?", "How much do AI auditors make?", "Will auditor jobs be replaced by AI?" — and then off the topic entirely, into "What did Stephen Hawking say about AI before he died?", "What did Nikola Tesla say about aliens?", and finally a run of stock-picking questions ending at "What are the 7 stocks to buy and hold forever?"

We recorded a related failure on 23 September 2026, when six of the nine People Also Ask questions on aeo agency turned out to be about Authorised Economic Operator customs status. That one at least had a real homonym behind it. This has nothing: Google reads "AEO audit", finds no entity, falls back to "AI", and drifts.

The practical consequence is that the category is cheap and unstable at the same time. Nobody has to outspend anyone to own this phrase today. But an answer assembled with no entity anchor is an answer that can change completely, and we have watched exactly that happen — the five agencies Google's AI named for one AEO query on 23 September had been replaced by four entirely different names twelve days later.

What should an audit cost, and what should it produce?

The published range on one page of results runs 30 to 1. Pixelmojo sells a single-page AEO audit for $5. Tryansly advertises "47 checks across 7 categories in 30 seconds", free. PitchKitchen offers a free audit "built by hand". Meanwhile advertisers bid $148.72 a click on seo audit agency — the highest cost per click we have measured anywhere in this category.

That spread is not a pricing mystery. It is two different products sharing a word. A $5 automated scan reports what a crawler can see. What it cannot do is the measurement half of this checklist — items 20 to 25 — because that needs a prompt set, a competitor set and a judgement about which gaps are worth closing.

A reasonable test before you pay for anything: ask what the deliverable contains beyond a score. A number between 0 and 100 is not a finding. A ranked list of blockers with an owner and an expected effect against each is. What an audit costs works through the ranges and what sits behind them.

Can you run this yourself?

Most of it, yes.

Groups one and two are a good afternoon's work with free tools and a text editor, and you should do them yourself whoever you eventually hire, because they are also the cheapest things to fix. Group three is slower but mostly editorial.

Group four is the one that resists self-service, and not because it is clever. It needs a fixed prompt set, repeated measurement, and the discipline to keep asking the same questions after the answers stop flattering you. The tools that measure it compares what is available, and how to track what AI engines say about you covers the method.

One honest caveat we publish about ourselves: at the last measurement, stefka.com had zero genuine AI mentions — one recorded mention, co-cited with a baby-name dictionary, which is about the given name rather than the studio. We are running this checklist on ourselves in public, and that is the number it currently returns.

Where Stefka fits

We are an AI-native marketing studio for B2B SaaS, run from Amsterdam, with AI search visibility as the specialism and fractional marketing leadership as the way it is bought.

An audit is usually where that starts, because the measurement half decides what the rest of the work should be. If you want the checklist run properly against your site and your competitors' answers, talk to us about an audit or read what an engagement costs first.

Frequently Asked Questions