MarqueeOS / Answers
Answer

Your agency says it does AI search.
What should you ask for?

Six questions to ask an agency that says it does AI search (AEO or GEO), what a good answer sounds like, and the red flags that mean you are paying for a label.

One questionUpdated 2026-11-09No signup

Ask an agency for six things: its question list, repeat checks, raw answers, listings plan, a dated change log, and a monthly report. In a 720-answer study, one check missed 75% of the companies AI names only some of the time, so a single screenshot proves little.

What AEO and GEO mean, and why the label tells you little

AEO, answer engine optimization, is the work of getting a business named when someone asks ChatGPT, Perplexity or Google's AI answers for a recommendation. GEO, generative engine optimization, is the same goal under a second name, and most agencies use the two interchangeably.

One trap: some vendors use GEO to mean geotargeting, which is showing different content or ads to people in different places. That is an old local marketing tactic and has nothing to do with AI answers. If an agency says GEO, ask which one it means. The answer tells you a lot before you ask anything else.

How AI search overlaps with local SEO

A lot of the work is shared with local SEO, and that is good news if you already pay for it. Your Google Business Profile, your listings on directories like Yelp and Bing Places, and your reviews feed both your Google page and the answers AI engines write about local businesses.

BrightLocal's directory study (published 2026-08-26, source) found Google Business Profile made up "28.5% of all citations" in local AI answers, and Yelp "appeared as a source in 80% of all answers from ChatGPT". So an agency that bills you twice for the same listings work, once as local SEO and once as AEO, deserves a question. What is different is the measurement. Local SEO tracks where you rank. AI search asks whether you get named at all, and that has to be checked by asking the engines.

1. Which questions are you tracking, and can I see the list?

Ask for the full list of questions, in writing, in the words a buyer would type. For a plumber that might be "who is the best emergency plumber in Austin" or "how much does it cost to repipe a house". A good answer is a list of 10 to 20 real buyer questions you can read, edit and add to, with the city in them.

A weak answer is a score with no questions behind it, or a list of keywords instead of questions. If you cannot see what was asked, you cannot tell whether the agency is measuring anything you care about.

2. How many times do you ask each one, and on which engines?

A good answer is a number and a list of engines: for example, each question asked eight times on each engine you care about, in fresh sessions. AI engines do not give the same answer twice, so one check is a coin flip.

Our 720-answer study asked six builder questions eight times each in 15 US metros on Perplexity in September 2026. 47% of named companies appeared in all 8 asks. 32% appeared in 4 or fewer, and a single check missed those companies 75% of the time. A company on the edge of the list looks invisible on one check and present on eight. An agency that screenshots one answer cannot tell which you are.

BrightLocal's study of about 200,000 prompts reported 23% overlap on ChatGPT repeat runs (published 2026-09-16, source), which points the same direction. Also ask which engines. ChatGPT, Perplexity, Gemini and Google's AI Overviews do not match each other, and an agency that checks only one should say so.

3. Can I see the raw answers, the prompt log?

The prompt log is every question asked, the date, the engine, and the answer word for word. Ask for it as a file or a shared sheet. A good answer is yes, and it arrives the same day.

A score or a chart can be wrong, and you cannot check it. The raw answer shows who the engine named instead of you, which sources it cited, and whether it got your business name, city or services wrong. That last one is often the most useful finding, and it only shows up if someone reads the answers. If the agency says the log is proprietary, treat that as a no.

4. Which listings and directories will you work on?

Ask for names, not "citations" in general. A good answer lists specific places: your Google Business Profile, Yelp, Bing Places, Apple Maps, the Better Business Bureau, and the directories your trade uses. It also says who owns each login. The BrightLocal figures above are why this matters. For local businesses, those listings are a big part of where the answers come from.

The thing to check is that your business name, address, phone number, hours and services match everywhere. If the agency's plan is mostly new blog posts and backlinks, ask how that connects to the places AI engines actually pull local answers from.

5. What did you change, and when?

A good answer is a dated change log: on this date we fixed the hours on Bing Places, on this date we added a page answering this question. With that log and the prompt log side by side, you can see whether an answer changed after a specific change.

Without dates, any improvement gets credited to the agency and any drop gets blamed on the engine. Even with dates, you will not get proof of cause, because the engines change on their own. You will get a record you can argue from, which is more than most owners are given.

6. What will you report monthly, and what can you not promise?

A good monthly report has the same questions run the same way every month, the share of answers that named you out of the total asked, who was named instead, and the change log. It should also say what the agency does not control. Nobody controls the answer an AI engine writes.

The honest promise is the work and the measurement: what gets done, by when, and how it is checked. It is not a position. Some timelines also run longer than a month, since engines re-read sites on their own schedule. Ask what the agency expects at 30, 60 and 90 days, and what it will say if nothing moves.

Red flags

A guaranteed ranking or placement in ChatGPT. Nobody can guarantee that.

A single screenshot offered as proof. One answer shows what happened once.

No list of questions, or a list you cannot see.

A score with no raw answers behind it.

Four months of backlinks and meta descriptions with no answers checked. That is ordinary SEO work, which may be worth paying for, but it is not a way to find out whether AI engines name you.

A refusal to say which engines were checked, or what "GEO" means in their proposal.

What this does not tell you

The 720-answer study covers builders and remodelers on Perplexity in September 2026, so it does not show how often other trades or engines vary. The BrightLocal figures describe local AI answers in general and not your market. A good answer to all six questions does not mean the agency will get you named. It means you will be able to see whether it did. For a shorter list aimed at AI visibility vendors, see the AI visibility page.

Check your own answer first

Before you hire anyone, run the free AI Visibility Check to see what the engines say about your business today.

Still the wrong
answer for you?

Thirty minutes, no deck. Bring one messy process and the list of systems it touches. If the honest answer is that you do not need us, you will hear it on the call.

Book a 30-minute call →