How to Choose an AEO Agency: Questions, Scorecard and Red Flags (2026)

The short answer

Pick the AEO agency whose numbers you can check yourself. Before you talk to anyone, run your own baseline. Then ask each agency what it found when it ran your prompts, ask for proof you can rerun, and score them all on the same weighted scorecard. Drop anyone who scores under 70 out of 100.

A disclosure first. I run AEO Labs, an AEO agency, so I sell the thing this page helps you buy. This page names no other agency, and the checklist works the same whether you hire us, someone else or nobody. If you want named options to compare, our list of the best AEO agencies covers the field.

This page is the hiring checklist. For what a full engagement should contain, see our guide to answer engine optimization services. For the choice between doing it yourself, hiring a consultant or buying a retainer, see answer engine optimization consulting.

Run your own baseline before any sales call

You will judge every pitch better if you already know where you stand. This takes an afternoon.

1. Write down about 10 prompts a buyer would type when shortlisting in your category: "best [category] for [use case]", "[competitor] alternatives", "[your brand] vs [competitor]", "is [your brand] worth it".
2. Run each one in ChatGPT, Perplexity, Gemini and Google AI Overviews. Use a private window or a logged-out session where you can, and note the date and the country you ran it from.
3. For each answer, log four things: whether your brand is mentioned, whether your site is cited as a source, which competitors are named, and which URLs the engine cited.
4. Run the same set again on a different day. Answers change between runs, so a prompt that names you once and not the second time is a coin flip.

That gives you about 40 answers per pass and a short list of the sources deciding your category. Our walkthrough on how to check if ChatGPT recommends your brand covers the manual method in more detail.

Now you have a test for every agency. A good one will have run something similar before the call and will tell you what it saw. If its findings contradict yours, ask why. Sometimes it ran more prompts or a different market. Sometimes it did not look.

12 questions to ask an AEO agency

Each question below comes with what a strong answer and a weak answer tend to sound like. You are listening for specifics about your brand.

1. What did you find when you ran our prompts?

A strong answer names prompts, engines and competitors from your category, and at least one source URL you are missing from. A weak answer is a general talk about how AI search is changing, with nothing that required looking at your brand.

2. Which prompts would you track, and where does the list come from?

Strong: the list starts from real buyer language, such as questions your sales team hears, review sites and community threads, then expands across the buying journey. They expect you to review and edit it. Weak: a list generated from keyword tools with no input from you, or a vague promise to track "thousands of prompts" with no say in which ones.

3. Which engines will you cover, and how does the work differ between them?

Strong: they name the engines your buyers use and explain that the engines retrieve differently. Google AI Overviews leans on what already ranks in classic search, while ChatGPT and Perplexity lean harder on retrieved third-party pages, so the effort is weighted per engine. Weak: "AI search" treated as one thing, or a single engine sold as the whole job.

4. What exactly do you measure, and how often?

Strong: mentions and citations tracked separately, per prompt and per engine, against named competitors, on a fixed schedule. They show you a sample report. Weak: a single proprietary visibility score they cannot break down, or traffic graphs alone.

5. How many runs does each number rest on?

Strong: they know answers vary between runs and sessions, and can tell you how many times each prompt is run and whether a reported share is an average. Weak: surprise at the question, or one run per prompt reported as fact.

6. Can I open your proof myself?

Strong: share links hosted on the engine's own domain, or the exact prompt, engine, date, location and session type so you can rerun it. Weak: cropped screenshots, logos on a slide, or percentages with no starting number.

7. How much of the work happens off our website?

This one matters most. When AI engines build an answer, most of the sources they cite are third-party pages: editorial roundups, community threads, review and reference sites. A brand's own website is usually a small share of what gets cited. You can see this in your own baseline test by sorting the cited URLs into yours and everyone else's.

Strong: a named plan for the sources your baseline turned up, such as roundups, review platforms, publications and community threads, with examples of placements they have earned for other clients. Weak: a plan that is all on-site content and markup.

8. Are placements earned or paid, and is paid spend inside the fee?

Strong: they separate earned and paid placements, show you the price of any paid one before buying, and wait for your approval. Weak: "we have relationships" with no list, or a fee that hides what was bought.

9. Who does the work, and who signs off content?

Strong: named people, how much of their week goes to your account, and a review step where your subject-matter experts approve anything that makes claims about your product. Weak: a senior person on the sales call and nobody you can name after it.

10. How will you work with our SEO agency, PR team and developers?

Strong: a clear split of who owns what, and a plan for the technical fixes your developers will need to ship. Weak: either "we handle everything" or no answer at all, which tends to end in duplicate work or a stalled backlog.

11. What will the first month produce?

Strong: concrete artifacts. A prompt list, a baseline with source URLs, a diagnosis of why you are missing, and a prioritized plan that splits on-site fixes from off-site targets. Weak: "onboarding" and "strategy" with nothing you could hand to another team.

12. What will you not promise?

Strong: a quick, specific answer. No guaranteed positions, no fixed timelines before an audit, nothing permanent, because answers are rebuilt on every query. Weak: a guarantee of any kind.

A weighted scorecard you can copy

Score each agency from 0 to 5 on each row, multiply by the weight, divide by 5, and add it up. The total is out of 100.

Pass threshold: 70. Below that, keep looking, however good the call felt. Above it, the highest score is your shortlist, and the references decide between close scores.

Weight proof and off-site work most because they are the hardest to fake on a call. Anyone can describe a method. Fewer agencies can hand you an answer you can open and a placement you can visit.

How to check an agency's proof yourself

AI answers change between runs, between logged-in and logged-out sessions, between countries and between model versions. A single screenshot shows what one person saw once. Treat it as a claim and check it.

Ask for an engine-hosted share link first. ChatGPT and Gemini can publish a conversation at a public link on their own domain, so you see the original answer without trusting an image the agency made. If there is no link, ask for five details: the exact prompt as typed, the engine, the date, the location, and whether the session was logged in. Then run it yourself.

The exact wording matters more than people expect. When we captured ChatGPT for one of our clients, Try Create, on August 9, 2026, a logged-out US session ranked it first for "what is the best creatine gummy". The plural, "best creatine gummies", returned an answer that named the brand without ranking it first. Both are true, which is why the prompt has to be quoted character for character.

Results also age. When we reran a 2025 ChatGPT result for another client in August 2026, it no longer reproduced, so we now date that case study as a 2025 result. Ask every agency when its proof was last checked.

Then ask for the baseline. A big percentage rise in citations means little without the starting count and the number of tracked prompts. Ask which tool produced the number and how many runs it rests on. If the agency cannot tell you, score the proof row low.

Red flags

• A guaranteed position in ChatGPT or any other engine, or a refund tied to one.
• A timeline quoted before anyone has looked at your prompts.
• Proof that is only screenshots, logos or percentages without a starting point.
• A plan that is all on-site content and markup, with no off-site work.
• Tracking you cannot see, or a score they will not break down.
• No answer when you ask what they found before the call.
• Pressure to sign a long term before any baseline exists.
• Vague talk about "relationships" with publishers and no list of placements.
• Contracts where the agency keeps the prompt list, the tracking history or the tool accounts.

Agency vs in-house vs your current SEO agency

The work needs five skills. Who has them decides your route.

When in-house wins

In-house is the better call when most of these are true:

• A named person owns AI visibility and has real hours for it every week.
• You have writers and subject-matter experts who can review claims fast.
• Someone on the team already earns press, reviews or placements.
• You can access your own analytics and CRM to see AI-referred visits and deals.
• You are willing to rerun the baseline on a schedule for months.

Can my SEO agency just do this?

Maybe. Ask it four things. Has it got a client named in an AI answer, and can it show the prompt, engine and date? What off-site placements did that take? How does it track AI citations separately from rankings? Who on its team does outreach? If the answers are all on-site work, you are buying SEO under a new name. That work still matters, especially for Google AI Overviews, but it covers a small slice of what the engines cite. Our AEO vs SEO guide covers where the two overlap.

The hybrid

Many teams split it. The agency runs prompt research, measurement and the off-site work, which are the parts that are hardest to staff. Your team writes or approves content and ships technical fixes. Your SEO agency keeps the site healthy. This works when one person on your side owns the whole picture and the agency reports to them.

Contract and data ownership questions

Ask these before you sign, and get the answers in writing:

1. Who owns the prompt list? It should be yours.
2. Can I export the full tracking history, per prompt and per engine, in a format I can open elsewhere?
3. Whose name is on the tracking tool account? If it is the agency's, what happens to the history when we leave?
4. Who owns the content you produce, including drafts?
5. What happens to paid placements after the engagement ends? Do they stay live?
6. What is the minimum term and the notice period?
7. Is there a fixed-scope first phase, with a defined deliverable, before any retainer?
8. What exactly is in the monthly fee, and what is billed on top?

A short first phase is a good sign. It lets both sides find out whether the work fits before either commits to months of it.

What to ask an agency's references

References are picked by the agency, so ask questions that are hard to answer with praise alone:

• What was your baseline when you started, and where is it now?
• Did the results hold after the engagement ended or slowed?
• Who did the work day to day, and was it the person you met on the sales call?
• What did the agency get wrong, and how did it handle a bad month?
• How much of your team's time did it take?
• Would you hire them again for the same job?

Ask for a reference whose engagement ended. A current client is still in the middle of it.

Key takeaways

• Run your own baseline of about 10 buying prompts across four engines before any sales call, and use it to grade every pitch.
• Ask what each agency found before the call. Specifics about your brand separate the agencies that looked from the ones that did not.
• Weight proof and off-site work most. Most of what engines cite sits on other people's sites, so an all on-site plan works on a small slice.
• Check proof yourself with engine-hosted links or the exact prompt, engine, date, location and session, and ask how many runs a number rests on.
• Score every agency out of 100 and drop anyone under 70.
• Choose in-house when you have the owner, reviewers and PR skills; choose an agency when you do not; consider a hybrid when you have some.
• Own your prompt list, tracking history and content in the contract.

Where AEO Labs fits

We are one of the agencies you could run this checklist on. Every engagement starts with a 30-day sprint scoped to your brand, and it ends with a baseline, a plan and the first fixes shipped. We do not have a fixed price list. We quote the sprint after a first call where we look at your category. Across the agencies we compared, published prices run from about $3,000 a month to $10,000 a month and up, so compare proposals on deliverables and on how results will be measured.

How we run the work is on our answer engine optimization agency page, and the engine-hosted links behind our client results are on the proof page. If you want to score us, book a call and ask us the 12 questions above.