How to Evaluate an AEO Agency in 2026: A 10-Point Test
Answer engine optimization is about eighteen months old as a category, which means the field contains genuine specialists and repackaged SEO in roughly equal measure. This is the rubric we would use, published in full, including the four tests you can run on any agency's own website in ten minutes.
Why this list exists, and how it was built
Most "best agency" lists do not say how they chose anyone. We checked the list currently ranking for the main Atlanta agency query: it gives a starting price, a founding year and a proof number for each entry, and discloses no selection method at all. That is not useless, but it is not a test you can apply yourself.
So this is a rubric rather than a leaderboard. Every criterion below is weighted by how strongly we think it predicts results, the weights are stated, and four of the ten can be verified from outside without talking to anyone. We have deliberately not published a ranked list of named competitors, because scoring other firms on evidence we cannot fully verify is not something we are willing to present as fact.
Disclosure: Engage Ads is an answer engine optimization agency. We wrote the rubric we would want a buyer to apply to us, which means we score well on some criteria and it is fair to weigh that when reading. Every test below can be run against engageads.io as easily as against anyone else.
The rubric at a glance
| # | Criterion | Weight | Verify from outside? |
|---|---|---|---|
| 1 | Their own site is readable by AI crawlers | 15 | Yes |
| 2 | Their structured data renders server-side | 15 | Yes |
| 3 | Their own entity facts are consistent | 10 | Yes |
| 4 | They are cited by assistants in their own market | 5 | Yes |
| 5 | They measure citation share properly | 15 | No |
| 6 | They work on off-site corroboration, not just pages | 15 | No |
| 7 | They publish original data | 10 | Yes |
| 8 | Pricing structure is flat, not spend-linked | 5 | Partly |
| 9 | You keep ownership of accounts and data | 5 | No |
| 10 | They refuse to guarantee placement | 5 | No |
Weights total 100. The two heaviest technical criteria are weighted equally with the two heaviest process criteria, on the reasoning that an agency which cannot do the technical work is useless, and an agency which does only the technical work plateaus after about three months.
Test 1: Fetch their own site the way an AI crawler does
Weight 15. Verifiable from outside.
Request their homepage with a GPTBot user agent and compare the response to what a normal browser gets. You are checking two things: that the request is not blocked, and that the content is materially the same. Sites that build their content client-side return a near-empty shell to these crawlers.
Pass condition: the AI-crawler fetch returns the same substantive content as a browser fetch, at a comparable size.
Test 2: Check their structured data is in the raw HTML
Weight 15. Verifiable from outside.
View source, not the inspector. The inspector shows you the page after JavaScript has run; view-source shows what a non-rendering crawler receives. Search for application/ld+json and confirm there is real schema inside, describing the organization, its services and its location.
This is the single most common failure. Plenty of agencies selling AEO have schema that only materialises after JavaScript executes, which means GPTBot, ClaudeBot and PerplexityBot never see it.
Pass condition: Organization or LocalBusiness schema, plus service and FAQ markup, present in view-source with real values in it.
Test 3: Compare their own entity facts
Weight 10. Verifiable from outside.
Collect the phone number and address from four places: their website footer, their structured data, their Google Business Profile and their LinkedIn page. Then check whether all four agree.
Inconsistent contact facts are the exact defect an AEO agency is hired to fix. Finding it on their own properties is disqualifying in a way that is hard to argue with, and it takes about three minutes.
Pass condition: one name, one address, one phone number, in all four places.
Test 4: Ask an assistant to recommend an agency in their city
Weight 5. Verifiable from outside.
Run the query their own buyers would run. Weight this lightly and read the result carefully, because a young agency doing correct work may not be cited yet, while an old agency with a large backlink profile may be cited for reasons unrelated to AEO skill.
The more informative move is to ask them why they are or are not cited. An agency that can explain its own position specifically, naming the sources the assistant used, understands the mechanism. One that is evasive does not.
Pass condition: they can explain their own citation position specifically, whether or not it is flattering.
Test 5: Ask exactly how they measure citation share
Weight 15. Ask them.
Generative answers vary between runs for the same prompt. Any measurement built on a single run is noise. A credible answer includes: a fixed panel of prompts, repeated runs, a result expressed as a rate such as named in six of ten runs, a stated minimum number of runs before a data point is trusted, and a rule that editing a prompt resets the trend.
Ask what happens to the trend line when they change a prompt. If they have not thought about it, they are not really measuring.
Pass condition: a fixed prompt panel, repeated runs, results as a rate, and an explicit minimum sample.
Test 6: Ask what they do off-site
Weight 15. Ask them.
This separates AEO from repackaged on-page SEO. When assistants rank service providers, they lean heavily on independent corroboration: directory profiles, verified reviews, industry listings, press. An agency whose entire plan happens on your website will improve your machine readability and then stall.
Pass condition: a concrete plan covering directory profiles, a sustainable review process, and getting into third-party listings.
Test 7: Check whether they publish original data
Weight 10. Verifiable from outside.
Original first-party data is the most-cited content type, because it is the only kind a competitor cannot reproduce. An agency that publishes benchmarks from its own campaigns, with sample sizes and periods attached, both demonstrates the technique and proves it has enough data to have learned something.
Check that third-party figures in their content carry a publisher, a sample size, a period and a link. Unsourced statistics in an agency's own marketing predict unsourced statistics in yours.
Pass condition: at least one original data asset, and sourced third-party figures elsewhere.
Test 8: Check the pricing structure, not the price
Weight 5. Partly verifiable.
A percentage of ad spend creates an incentive to spend more of your money. It is also incoherent for AEO specifically, which has no media spend to take a percentage of. A flat monthly fee is the structure that matches the work.
Published pricing is a mild positive on its own, since it correlates with willingness to be measured.
Pass condition: flat monthly fee, disclosed before a sales call, with no long-term lock-in.
Test 9: Confirm you keep ownership
Weight 5. Ask them.
Ad accounts, analytics, creative assets, CRM and any schema or content built for you should be yours, in your accounts, and should survive the engagement ending. Ask specifically what happens on day one after you leave.
Pass condition: everything is in accounts you own, and you keep it all.
Test 10: See whether they refuse to guarantee placement
Weight 5. Ask them.
Nobody controls what a generative model returns, and retrieval behaviour changes when models are updated. An agency that guarantees you will appear in ChatGPT or AI Overviews is either misunderstanding the mechanism or knowingly overselling. The right answer is a guarantee about inputs and measurement, never about placement.
Pass condition: they decline to guarantee placement, and can explain why without being cagey.
How to score it
Award the full weight where the pass condition is met, half where it is partly met, zero where it is not.
| Score | Reading |
|---|---|
| 80 to 100 | Genuine specialist. Expect the technical work to be right and the off-site work to be real. |
| 60 to 79 | Competent, probably strong on one half and thin on the other. Ask directly about whichever half scored low. |
| 40 to 59 | Likely an SEO agency that has added AEO to the menu. Can still be useful, but price it accordingly. |
| Below 40 | Selling a label. Failing tests 1 to 3 on their own website puts a firm here regardless of the rest. |
One weighting note worth stating: tests 1, 2 and 3 are all verifiable from outside and total 40 points. If an agency fails all three on its own domain, the remaining 60 points are not worth collecting.
Straight disqualifiers
Four things should end the conversation regardless of everything else.
- A guarantee of AI placement. It is not achievable, so it is either ignorance or overselling.
- Their own site fails the crawler test. If GPTBot cannot read their homepage, or their schema only exists after JavaScript runs, they do not practise what they sell.
- They hold your accounts. Any structure where leaving means losing your ad account, your data or your content is a commercial trap independent of the quality of the work.
- Unsourced statistics in their own marketing. An agency that publishes numbers without a publisher, sample or period is telling you what its reporting to you will look like.
Frequently asked questions
What is the difference between AEO and GEO?
None that matters in practice. Both describe making a business readable, verifiable and quotable to AI assistants. Some practitioners use GEO for the generative answer layer and AEO for featured snippets and voice search, but the techniques overlap almost entirely and no standards body distinguishes them.
How much should AEO cost in 2026?
Most credible engagements for a local service business fall between roughly $1,000 and $5,000 per month, depending on how much entity and structured-data remediation is needed up front. Be more suspicious of the structure than the level. A flat fee is the honest shape for work with no media spend attached.
Can I do AEO myself?
The audit, yes. Tests 1, 2 and 3 work on your own site as well as on an agency's, and finding conflicting phone numbers or client-side-only schema is genuinely something you can do in an afternoon. The parts that are hard to self-serve are sustaining a review process and building the off-site corroboration layer over months.
How long before AEO shows results?
Technical fixes are readable by crawlers within days. Corroboration takes weeks to months because it depends on third parties. Treat anyone promising a specific week as guessing, and ask instead how they will show you the trend.
Run the first four tests on us
Fetch engageads.io as GPTBot, view our source, compare our phone number across four places, and ask an assistant about agencies in Atlanta. Then book the free growth audit and we will run the same tests on your business and give you the findings in writing.
Related reading: straight answers to common marketing questions, the marketing and AI search glossary, and our 2026 Georgia advertising benchmark built from first-party campaign data.
Related: best AEO agencies for small businesses, 10 compared.
Related: how ChatGPT, Perplexity and Google AI choose a marketing agency: a 95-check study and GEO for local businesses.
Want to check the agency's claims against the actual mechanics? Our guide to getting recommended by ChatGPT, Gemini and other AI assistants lists what each engine reads and the 30-day order of operations.
If you would rather see how we answer these ten questions ourselves, here is what our AEO service actually covers.
