SEOAIO.ai

Buyer’s guide

How to hire an AEO agency without paying for a screenshot

Agencies now sell answer engine optimization as a monthly service, often to buyers who have no way to check it. These are four questions to ask before you sign, and what to do after, so the work is judged by something other than the people doing it.

We name nobody here. Every point below came from reading real agencies' pages, but a page can change next week, and the method is worth more than a list of names that goes stale.

1

Get the target questions in writing, before the work starts

Ask: Which questions, asked by which buyers, will this engagement move?

  • Ask for the list of questions the agency is working on, in the words a buyer would type. Five named questions are a plan; "your AI visibility" is not.
  • Get it before the first piece of work goes live. A list written after the results come in can simply be the questions that happened to move.
  • Keep the questions they did not pick. Those are your comparison: if everything rises together, the market moved, not their work.

The tell

An agency that will talk about outcomes but not name the questions it is working on.

Vendor Check records the agency's questions against the ones they left alone, before the next measurement, so the split cannot be chosen after the fact.

2

Ask how they measure, down to the count

Ask: How many times is each question asked, on which engines, and what counts as appearing?

  • Ask how many answers each result is based on. The same question asked twice can name different businesses, so one answer is one sample, not a finding.
  • Ask which engines, by name, and whether each is measured or only mentioned in the pitch.
  • Ask what counts: being named, being linked, or being the first recommendation. They are different results and they move differently.
  • Ask for the denominator beside every percentage. A share with no count under it cannot be checked.

The tell

A screenshot of one chat answer offered as proof, or "we checked and you are there".

Why asking an assistant once yourself is not a measurement, compared side by side with sampling the same question many times.

3

Ask for one result dated after they started selling this

Ask: Can you show a result from an AEO engagement, with the start date, the window and the count?

  • Check the dates on their case studies against when the AEO service appeared on their site. Results from classic SEO work are real, but they are not evidence for this service.
  • Ask for the starting value. A rise with no baseline can be two answers becoming four.
  • Treat a timeline as a promise, not a result. An outcome stated as a number of weeks is a forecast unless someone measured it.

The tell

Case studies that all predate the service, or percentages with no starting point.

The Vendor Claim Ledger grades, per vendor, what the marketing claims against what the entry price actually buys, dated and sourced.

4

Check that every surface they name still exists under that name

Ask: Which AI surfaces will you test on, exactly, and what are they called today?

  • Write down every product name in the proposal and search each with the word renamed or retired. AI search products are renamed often, and a proposal written from an old playbook shows it.
  • Be wary of rich-result work sold as AI visibility. Some structured-data rich results that proposals still sell have been withdrawn by Google.
  • Ask whether "getting into the model's training data" is part of the plan. Nobody outside the model's maker can do that on request.

The tell

A test platform or result type that stopped existing under that name before the proposal was written.

We do not have a tool for this one yet, so it is the manual check above. The engines we measure are listed by the names they carry today.

After you hire: grade the work yourself

Once the engagement starts, the agency's report is one view. Yours can be another. Vendor Check lists the posts that went up on your site and how many measured answers cited each, counts answers citing the pages they placed on other sites, compares the pages they say they changed against the ones they did not touch, and grades their target questions against the rest. Every count carries the number of answers behind it, and none of it claims the agency caused a change: it shows what moved, and what did not.

Run the same questions on us

We are not an agency, and we do not sell AEO work. But a guide that exempts its author is an advertisement, so, in order:

  1. 1. You choose the questions we measure, and you can see and edit the list at any time.
  2. 2. After sign-up we sample real answers on Perplexity, Google AI Mode and AI Overviews, ChatGPT and Claude, and every share carries its count and a 95% interval. The free audit is different and says so: it reads your own pages and asks no engine anything.
  3. 3. We do not publish case studies without a starting value, a window and a count, which is why our proof page is short.
  4. 4. The engines above are named by what they are called today, from the same list our measurement runs on.

Run the free audit on your site · Checking a tool instead of an agency · What we do instead of writing to your site