12 questions to ask an AEO agency before you sign

Pedro Sena

Founder, Noudar AI

Short answer

Ask how they pick prompts, how they measure AI answers, what proof they can show with a baseline and timeframe, who does the work, what they will not do, and what you keep if you leave. Walk away from promised citations, schema or llms.txt sold as the fix, and undisclosed Reddit posting.

Key points

Ask how they pick buyer questions, how they measure AI answers, who does the work and what you keep if you leave.
Google says no special optimisation is needed for AI Overviews, so schema or llms.txt sold as the fix is a red flag.
Check every case study for a baseline, a time window, a method, a named client and a pipeline number.
In the contract, watch for promised citations, long lock-ins with no review point and assets owned by the agency.

Your buyers now ask AI what to buy. In a G2 survey, 51% of B2B software buyers start research in an AI chatbot more often than Google. So more SaaS teams are hiring AEO agencies. Most of those pitches sound alike.

This page is the checklist for the sales call. Take it in, ask the 12 questions, and listen for the red flags. It is written for a B2B SaaS team between about 1M and 10M ARR: one budget, a small team, and no room for a wasted quarter.

For the full hiring process (brief, shortlist, prices and a 90-day test), read how to choose an AEO agency.

What are the 12 questions, and what should you hear?

Ask every agency the same 12 questions, in the same order. A good answer is specific and written down. A red flag is vague, or a promise no one can keep.

#	Question	Good answer	Red flag
1	Which buyer questions will you track, and can I see the list before we sign?	A draft list built from your sales calls, demos, reviews and search data. You can edit it.	"We track thousands of keywords." No list until after you sign.
2	AI answers change from one ask to the next. How do you measure that?	Re-checks every question over time and reports ranges and trends, not one result.	One screenshot of one good answer as proof.
3	Which AI engines do you track, and do you report them apart?	Names each engine your buyers use and reports each one on its own.	One blended "AI visibility score" for everything.
4	What does Google say about AI search, and how does your plan fit it?	Knows Google's own docs. Treats AI search as SEO done well.	Special AI markup, llms.txt or "chunking" sold as the main lever.
5	Can you show a case study with a baseline, a time window and a pipeline number?	Yes, and it says how the result was measured.	A big percentage with no start point, dates or method.
6	How will you connect AI answers to demos and pipeline?	An AI channel in analytics, a "how did you hear about us" field and a CRM check.	Traffic and visibility only. No word on leads.
7	Who will run my account at month six, by name?	A named person and their role. You meet them before you sign.	The pitch team, then someone you never met.
8	Where do you use AI, and who checks its work?	Says what AI drafts and which expert reviews it before it ships.	Page volume as the selling point. No review step.
9	How will you get us mentioned on other sites and in communities?	Earned mentions and disclosed accounts only, inside each community's rules.	Posting from fake or undisclosed accounts. Paid or made-up reviews.
10	What do you need from us each week?	A named approver and a set amount of time for reviews, sales input and site changes.	"Nothing, we handle it all." Content written without your expertise.
11	What will you not do?	Names its limits, the work it turns down and the promises it will not make.	"We do everything."
12	What do we own if we leave?	All content, data, the question list and every account, in your name.	Accounts, tools or content held in the agency's name.

Three of these need a little more context.

On question 2. AI answers move a lot. From our platform, only about 1 in 8 of the companies named (12%) appeared in both ChatGPT's and Perplexity's answers. One answer on one engine proves very little. Ask what sits behind any chart they show you.

On question 4. Use Google's own words as the test. Google says no extra requirements or special optimisations are needed to appear in AI Overviews and AI Mode. Its AI search guide says optimising for generative AI search is still SEO, with no special schema, AI text files (llms.txt), Markdown or content chunking needed. Schema and llms.txt are hygiene. They are fine to do. They are not why AI names you. llms.txt is still a proposal by Jeremy Howard (2024).

On question 8. AI-written pages at volume carry risk. Google warns against scaled content abuse and asks for helpful, people-first content. AI should speed up the work. An expert should still own every page.

How do you check a case study?

Run every case study through five checks. If it fails two or more, treat the result as a claim, not proof.

Baseline. Where did the client start? "Up 400%" means little if the start was 5 visits.
Time window. Over which dates? Does the window start when the work started?
Metric and method. What was measured, with which tool, and how often? For AI answers: which questions and engines?
Named client. Is the client named, and can you speak to them? An anonymous case is weaker, though NDAs are real.
Pipeline. Does it report leads, signups, demos or revenue? Or only traffic and visibility?

Most published proof stops at visibility. In our read of 52 published agency case studies, 26 were SEO or AEO engagements. Only 10 of those 26 passed all five checks above. All of those results are the agencies' own claims, self-reported and not checked by anyone else.

Two more things to watch:

Small bases make big multipliers. A jump from near zero looks huge. Ask for the absolute numbers too.
The peak is not the result. AI visibility often spikes, then settles lower. Ask for the level it settled at, not the best week.

Which contract terms are red flags?

Read the contract before the call ends. These terms should make you stop:

Promised rankings or citations. No one controls what an AI engine says. A promise here means the agency either does not understand the work or is selling to you.
A long lock-in with no review point. Commit to a defined first phase with written goals, then decide. A 12-month term with no exit at all puts all the risk on you.
Tool fees passed through without detail. If you pay for tracking or content tools through the agency, ask for each tool, its cost and who owns the account.
No exit terms. The contract should state the notice period and what you get on the way out.
The agency owns your assets. Content, analytics, the question list and every profile should sit in your name from day one.
No written deliverables. If the contract lists hours or "strategy" but not pages, fixes and reports per month, you cannot hold anyone to it.
No disclosure rule for community work. The contract should say any mention or community work is disclosed.

For what agencies charge and how their terms compare, see how much an AEO agency costs.

Printable checklist

Copy this list into your notes and tick each line on the call.

Before the call

☐ Send the same one-page brief to every agency.
☐ Ask ChatGPT and Perplexity your top buyer questions. Note who gets named.

On the call

☐ They will share a draft list of buyer questions before you sign.
☐ They re-check AI answers over time and report ranges, by engine.
☐ They know Google's docs and do not sell schema or llms.txt as the fix.
☐ They show a case study with a baseline, dates and a pipeline number.
☐ They explain how they tie AI answers to demos in your CRM.
☐ You know who runs your account at month six.
☐ They say where AI is used and who reviews it.
☐ Community work is disclosed. No fake accounts or reviews.
☐ They tell you what they need from your team each week.
☐ They name what they will not do.

In the contract

☐ No promised rankings or citations.
☐ A first phase with written goals and a review point.
☐ Tool costs listed, accounts in your name.
☐ Clear notice period and exit terms.
☐ You own content, data and accounts.
☐ Deliverables per month written down.

Before you hire anyone, it helps to know where you stand.

Where does Noudar AI fit?

We turn AI visibility into pipeline. Noudar AI is an AI-native SEO and AEO agency for B2B SaaS and tech companies, based in Portugal. AI does the research, checks, first drafts and tracking. SEO and AEO experts make every call.

We are new and have no client case studies yet. So we cannot pass question 5 today, and we say so. Instead we run a Design-Partner batch: 3 spots, open until 31 October 2026. It costs EUR 1,500 at the start and EUR 1,500 at day 90. If we miss the goals agreed at kickoff, you get a full refund and keep the work. A case study with us is part of the deal.

Ask us all 12 questions. Hold us to the same answers.

Get a free AI visibility audit, results in 12 hours.

FAQ

Do the same questions work for a GEO or AI SEO agency?

Yes. GEO, AEO and AI SEO are different names for the same goal: getting your brand named in AI answers. Ask any of them the same 12 questions. If an agency says its label needs a different test, ask why. The proof standard does not change: a baseline, a time window, a method and a pipeline number.

Is it a red flag if an AEO agency has no case studies yet?

Not on its own. AEO is a young field, and new agencies start somewhere. The red flag is hiding it, or showing results with no baseline. A new agency should say so plainly, show its method in writing and share the risk, for example with a short first phase, clear goals and a refund if it misses them.

Who from our team should join the call with an AEO agency?

Bring the person who owns pipeline and the person who will approve the work. At a 1M to 10M ARR SaaS, that is often the CEO or Head of Marketing, plus whoever can change the website. If you can, add someone from sales. They know the questions buyers ask before a demo.

Should we start with a short test before a long contract?

Yes, if you can. A first phase of about 90 days with written goals shows how the agency works before you commit to more. Split the goals into work the agency controls, like pages shipped and fixes done, and results it can only influence, like AI mentions and demos. Then review together and decide.