Many AEO audit checklists start with the website: crawl access, schema, page speed, headings. That order works if you own one brand and one site. It is the wrong order for an agency. A client is unlikely to ask “is our schema valid?” The question is more likely to be “why does ChatGPT recommend our competitor and not us?” A crawl cannot answer that.
An AEO audit checklist for agencies runs in this order:
- Build the buyer prompt set.
- Run the prompts on each AI engine.
- Record mentions vs citations.
- Check entity signals and brand consistency.
- Check page structure and schema.
- Map the citation sources the engines trust.
Start with what buyers ask and what each engine says today. Work back to the site last. Below is each check in the order we recommend running it, with what to check and what good looks like.
What an AEO audit is (and how it differs from an SEO audit)
An AEO audit measures whether AI answer engines (ChatGPT, Gemini, Perplexity, and Google AI Overviews) name or cite a business when buyers ask questions it should win, then traces every gap to a fixable cause.
An SEO audit asks whether a page can rank. An AEO audit asks whether a business gets named in an answer. That depends on more than the client’s own pages, because engines can lean on third-party lists, directories, reviews, and forums to decide who to recommend. A site with clean technical SEO can still be absent from every AI answer in its category.
| SEO audit | AEO audit | |
|---|---|---|
| Unit of measurement | Keyword and ranking position | Buyer prompt and engine response |
| Main evidence | Crawl data, rankings, backlinks | Recorded answers, named and cited sources |
| Where fixes land | Mostly the client site | Client site, entity profiles, and third-party sources |
| How results move | Positions, fairly stable | Answers vary between runs, so the trend matters more than any one result |
The AEO audit checklist
1. Build the buyer prompt set
What to check: 10–20 prompts per client, phrased the way buyers ask an assistant. Cover four types: service (“emergency plumber near me”), comparison (“best hot water installer in Parramatta”), problem (“why is my drain gurgling”), and cost (“blocked drain repair cost Sydney”).
What good looks like: The set is agreed on with the account team before testing, contains no brand names, and maps every prompt to a service that earns the client money. Brand-name prompts prove nothing, because an engine asked about a business by name will usually describe it.
Agency note: use one prompt set template across all accounts, with the same prompt types, count range, and naming. That is what lets an SEO lead compare a roofing client in Calgary with a dental client in Brisbane without rebuilding the method each time.
2. Run the engines
What to check: Run every prompt on ChatGPT, Gemini, Perplexity, and Google AI Overviews, from a dedicated test profile with history and personalization turned off and the location set to the client’s market. Results can follow the tester’s IP address as well as the settings, so test from the client’s market or through a connection located there. Date every run, and run each prompt more than once in the same window, because answers vary.
What good looks like: Each engine is tested separately, because each can draw on different sources. A client cited in Perplexity may be absent in Gemini, and a fix that moves one engine may not move another. Never test from a personal account: saved history can skew answers toward what the tester has already searched.
Run the baseline by hand so you read the full answers. For monthly re-tests across many accounts, a tracking tool can carry the load, as long as it runs the same fixed prompt set.
3. Record mentions vs citations
What to check: For every prompt and engine, record the result: named (the business appears in the answer text), cited (its site is linked as a source), or absent. When a business is both named and cited, record both. Record which other businesses were named and which sources the engine cited.
What good looks like: A prompt visibility table with a date on every row.
Sample layout. Fictional client: Harbourline Plumbing, Sydney.
| Buyer prompt | ChatGPT | Gemini | Perplexity | AI Overview | Sources the engines cited |
|---|---|---|---|---|---|
| emergency plumber Parramatta | Named | Absent | Cited | Absent | Client site, two directories, one review site |
| best plumber for strata buildings | Absent | Absent | Absent | Absent | A best-of list, a forum thread |
| blocked drain repair cost Sydney | Absent | Absent | Cited | Named | Client pricing page, one competitor |
Top gap in this sample: no engine names the client for strata prompts, and the sources cited instead are third-party sources the client is not on.
The “sources cited” column is the most useful one in the audit. It shows where the engines get their answers, which is where the fix work goes.
4. Entity signals and brand consistency
What to check: Compare how the business is described on its site, its Google Business Profile, and the main directories and review sites in its market: name, primary category, service area, and core services. A local citation tool speeds this up across accounts, but read the five listings that rank highest for the client’s category by hand; the tool will not tell you which ones the engines actually cite.
What good looks like: The same name, category, and location everywhere. Engines can draw on many sources to work out what a business is and where it operates. When the website says “plumbing and gas,” the profile says “plumber,” and a directory lists an old suburb, the engine has three conflicting answers and may name a competitor whose signals agree.
5. Page structure and schema
What to check: For each prompt where the client is absent, find the page that should answer it. Does it answer the question in the first two to three sentences, use headings that match how buyers ask, and state quotable facts: services, area, process, and what the client actually does?
Then check structured data: Organization or LocalBusiness markup whose name, address, and service area match check 4, and FAQ content on pages that carry real buyer questions. Confirm robots.txt does not block the crawlers AI search features use, such as OAI-SearchBot (ChatGPT search) and PerplexityBot. Check Google-Extended too. It is a robots.txt token: blocking it does not affect Google Search, but it controls whether content is used for Gemini training and grounding in Gemini Apps. To appear in AI Overviews, a page must be indexed and eligible to show with a snippet in Search.
What good looks like: Each buyer question has one obvious page, that page opens with the answer, and the schema says what the page says. Schema does not rescue a page that never answers the question.
6. Citation sources
What to check: From the cited-sources column in check 3, list every third-party source that appears more than once across the client’s prompts: directories, best-of lists, review platforms, local publications, and Reddit or forum threads. Check whether the client appears on each and how it is described.
What good looks like: The client is present and accurately described on the sources the engines already cite for its category. Many site-first checklists skip this step, and for local service businesses it can hold the biggest gaps. A client cannot be recommended from a list it is not on.
Turning findings into a 30 and 90 day plan
An audit that ends with a list of problems is not finished. Trace every absent result to one of five causes: page answers, structured data, entity consistency, third-party citation sources, or reviews and profile signals. Then:
- Score each fix on expected impact and effort.
- Put fast, high-impact fixes on the 30 day list, such as entity corrections, page answer blocks, and schema.
- Put slower work on the 90 day list, such as earning places on third-party lists and building review depth.
- Give every action an owner type: content, technical, or off-site.
- Re-run the same prompt set each month and report the trend across the set.
The monthly re-test is what turns a one-off audit into something an account manager can report on. For how to track and present it, see how to measure AI visibility across client accounts.
FAQ
What is an AEO audit?
An AEO audit checks whether AI answer engines name or cite a business for the questions its buyers ask, records the result by engine and date, and traces each gap to a cause the business can fix.
How is an AEO audit different from an SEO audit?
An SEO audit measures rankings and site health. An AEO audit measures whether the business appears inside AI answers, and it looks past the site to the directories, lists, reviews, and forums the engines cite.
How many prompts should an AEO audit use?
For one local service business in one market, 10–20 buyer prompts show the patterns without turning the audit into a research project. Multi-market clients need a separate set per market.
How often should an AEO audit be re-run?
Monthly, on the same prompts. AI answers change between runs, so one result is not reliable; the trend across the full set is what shows progress.
Can an agency guarantee AI citations after an audit?
No. No one controls what an engine says. An audit identifies the gaps and the fixes, and progress is reported as a trend, not a guaranteed placement.
Run it on one client first
If a client is already asking why they do not show up in AI answers, test the method on that account. SearchKata runs a white-label AEO audit on one client site, in your template, at a fixed price: the prompt set, the engine results, and a ranked 30 and 90 day plan your team implements. See how white-label AEO for agencies works.
Leave a Reply