How to Choose an AEO Agency: 12 Questions to Ask
Choose an AEO agency by how it measures and what it changes, not by what it promises. Ask which assistants it checks, which buyer questions it uses and who wrote them, how many times it repeats each question, whether it separates mentions from citations, and how it reports shipped work apart from results. Ask what it will change on your site and who approves it. Be wary of guaranteed AI citations, claims of access to Google data, llms.txt sold as a Google ranking factor, and FAQ schema sold as a route to rich results: Google says no third-party tool has its internal data, that llms.txt does not affect Google Search, and it stopped showing FAQ rich results on 7 May 2026.
AEO (answer engine optimization), also called GEO or LLM SEO, is the work of getting a brand named and cited in answers from ChatGPT, Claude, Gemini, Perplexity, Grok and Google's AI features. Google's own guide says that, for Google Search, optimizing for its AI features is still SEO, and it published guidance in June 2026 on judging third-party AEO and GEO advice. That makes the buying question simple: can the agency show what assistants say about you today, change the right things, and report honestly on what moved?
What an AEO agency should actually do
- Measure: ask the assistants your buyers use the questions they ask, repeatedly, and record who is named and which sources are cited.
- Diagnose: check whether answer engines can reach and read your pages (robots.txt, CDN rules, rendering), and whether the pages answer the question in the first lines.
- Change: rewrite or create pages, fix technical blockers, and earn mentions on the third-party sources assistants cite, with your approval before anything ships.
- Report: show shipped work with dates, the measured change, and the normal variation of the numbers, separately.
12 questions to ask before you sign
Measurement
- Which assistants will you check, and on which surfaces? ChatGPT, Claude, Gemini, Perplexity, Grok, Google AI Overviews and AI Mode often disagree, so a single-engine number is a partial view.
- Which questions will you track, and who writes them? They should be your buyers' real questions, agreed with you, not your keyword list reworded.
- How many times do you run each question, and how do you show normal variation? The same prompt can return a different answer on the next run.
- Do you separate a mention (the brand named) from a citation (your URL listed as a source)? They move for different reasons.
- What is the baseline, and on what date was it recorded? Without a dated baseline, any later number is an anecdote.
Work
- What will you change on our site in the first 30 days, and who approves each change before it goes live?
- How do you check that answer engines can retrieve our pages? Ask for the actual test, such as fetching pages with each crawler's user agent and reading the response.
- What do you do off-site? Assistants cite third-party sources; ask how the agency earns real mentions, and how it avoids the inauthentic mentions Google's guide warns against.
- Will you create many near-duplicate pages for every question variation? Google's guide says doing that to manipulate AI responses violates its scaled content abuse policy.
Reporting and contract
- How will you report shipped work separately from outcomes, so we can see what was done even in a month where the numbers do not move?
- Which data comes from Google itself? Search Console's generative AI performance report shows your impressions in AI Overviews and AI Mode; ask whether the agency uses it.
- What do we keep if we leave: the question set, the raw answers, the change log and the drafts?
Red flags
| Claim | Why it is a red flag |
|---|---|
| We guarantee you will be cited by ChatGPT or Google | Google says third-party tools cannot guarantee performance, and indexing and serving are never guaranteed. No agency controls an assistant's answer. |
| We use Google's internal ranking data | Google says no third-party tool has access to its internal ranking or AI systems. |
| An llms.txt file will get you into Google's AI answers | Google said on 15 June 2026 that llms.txt files are not needed for Google Search and will not positively or negatively affect visibility or rankings. They may still help other services. |
| FAQ schema will win rich results | Google stopped showing FAQ rich results on 7 May 2026. FAQPage markup is still valid, but it is not a route to a rich result. |
| We will block or unblock the Google-Extended crawler | Google-Extended is a robots.txt token, not a crawler. It controls use of content for Gemini and does not affect Google Search. |
| You need special AI markup or chunked content | Google's guide says there is no special schema or file needed for its AI features and no requirement to break content into small pieces. |
| We will create a page for every possible question | Google's guide names this as scaled content abuse when done to manipulate AI responses. |
How to compare quotes
Agencies price AEO as a retainer, a project or an audit, and the names of packages vary, so compare what is inside them. Put each proposal into the same grid: number of tracked questions, assistants covered, runs per question, reporting frequency, pages written or rewritten per month, technical fixes included, off-site work included, and who owns the data. A cheaper retainer that tracks ten questions on one assistant once a month is a different product from one that runs fifty questions across five assistants weekly.
| Compare | Ask for a number |
|---|---|
| Questions tracked | How many, and agreed with you |
| Assistants covered | Which ones, by name |
| Runs per question | How many per reporting period |
| Pages per month | Written, rewritten or reviewed |
| Technical fixes | Included or billed separately |
| Data ownership | Raw answers and change log handed over |
Doing it in-house instead
Some teams hire an agency for strategy and keep measurement and fixes in-house. Fenn is built for that split. It asks ChatGPT, Claude, Gemini, Perplexity and Grok your buyers' questions when you start a check, reports share of answers, share of voice, average position, sentiment, competitors and cited sources, runs a readiness check on your main page, and turns each failure into a fix with a ready prompt. It drafts answer-first pages of 500 to 800 words and flags facts only you can supply. Every change waits for approval, and you publish it. Agencies use the same product for clients on the Agency plan, $199 a month for up to 25 websites. Accounts are set up after a demo.
FAQ
What does an AEO agency do?
It measures how AI assistants and Google's AI features describe and cite a brand, fixes what stops pages being retrieved or quoted, creates or rewrites pages that answer buyers' questions, works on the third-party sources assistants cite, and reports what changed.
Is AEO different from SEO?
For Google Search, Google says optimizing for its AI features is still SEO, with no extra requirements beyond being indexed and eligible for a snippet. Other assistants use their own crawlers and sources, which is why AEO adds measurement across several engines.
What should an AEO agency report every month?
The work shipped, with dates; the measured share of answers and citations per assistant against a dated baseline; the normal variation of those numbers; competitors named instead; and Search Console's generative AI impressions for Google.
Can an agency guarantee AI citations?
No. Google says third-party services cannot guarantee performance and that indexing and serving are not guaranteed, and other assistants change their answers between runs. Treat any guarantee as a red flag.
Put this to work on your own website.
Fenn finds what your customers ask, drafts the articles and site fixes, and measures what ChatGPT, Claude, Gemini, Perplexity and Grok say about you — with every change waiting for your approval.