How do I choose an AEO agency?

Last updated October 14, 2026. Written by the Third Coast AI team.

Choose an AEO agency the way you would choose any vendor whose work you cannot inspect directly: ask to see the prompt set they would measure you on, ask how many times they run it and in which engines, ask which pages the engines currently cite for your category and whether anyone on their team has read them, ask who writes and who reviews the content, and ask how the monthly report shows change at the prompt level rather than as a single score. An agency that can answer those five questions with examples from a real engagement is doing the work, and one that answers with a platform screenshot is reselling a subscription.

We wrote this page because the term now covers everything from a solo consultant with a spreadsheet to a software company with a services arm, and the price range and the quality range are both wide. It is written for the owner or marketing director comparing two or three proposals, and it applies to us as much as to anyone else. The pricing guide covers what the work costs; this page covers how to tell who should do it.

Which parts of the method should you ask to see?

Every credible AEO agency we have met runs some version of the same loop, which is to find out what the engines say now, work out why, change the pages and the third-party record, and run the questions again. The differences are in how carefully each step is done and whether you get to see it. The prompt set is the place to start because it is the measurement everything else rests on, and a prompt set written as keywords with "best" in front of them will measure keyword visibility and report it as AI visibility. Ask to see twenty prompts before you sign and read them as if you were the buyer; if they sound like something a customer would say on a call, including the awkward ones, the agency has done the thinking.

The second thing to see is a sample of the raw answers with citations, because the answers are where the plan comes from. Ahrefs measured the overlap between Google's top ten and the pages cited in AI Overviews at 38% in March 2026, down from 76% the previous July, so an agency that works from your rankings rather than from the engines' actual citations is working from the wrong list. Ask which pages the engines cite for your three most important questions and whether anyone has opened them; the answer tells you whether the content plan will be modeled on what the engines already reward or on a template.

Does the agency measure the way the research says to?

This is the question most buyers skip, and it is the one that decides whether the monthly report means anything. SparkToro and Gumshoe.ai had 600 volunteers run 12 prompts through ChatGPT, Claude and Google's AI a combined 2,961 times in January 2026 and found less than a 1 in 100 chance of getting the same list of brands twice, and about a 1 in 1,000 chance of getting the same order. That means any agency reporting your "rank" in AI answers is reporting noise, and the honest measure is the share of runs in which you are named, taken over dozens of prompts run several times each. Ask how many runs sit behind each number in the sample report; if the answer is one, the report is a snapshot and the trend line will be invented.

The same study found that visibility rates for the most-mentioned brands in a category ranged from 55% to 77% for headphones down to the 30s and 40s for design agencies, which is a useful reminder that the ceiling differs by category and that an agency promising you "number one in ChatGPT" is promising something the engines do not offer. We would rather hear a vendor say "we will move you from named in 12% of runs to named in 40%, and here is how we will know" than hear a guarantee. The measurement guide goes through the method in detail.

Should the agency write the content, or should you?

Either can work, and the choice depends on who knows the business. The pages that get cited answer the buyer's question in the first 150 to 200 words, state and prove the decision factors the engines use, carry statistics with sources and an attributed quote, and sit in the raw HTML where a crawler can read them. Aggarwal and colleagues at Princeton measured quotations, statistics and cited sources lifting a page's visibility in generated answers by about 41%, 33% and 28%, and keyword stuffing lowering it by 9%, so the craft is specific and learnable. What an outside writer cannot supply is the proof: the instructor ratio, the lead time, the warranty term, the thing your sales team says on every call. If the agency writes, make sure a reviewer on your side reads every page before it ships; if you write, make sure the agency supplies a brief built from the cited pages and runs a gate before publication. What should worry you is an agency that promises volume, because Glenn Gabe documented a site that lost visibility after publishing 850,000 AI-generated pages, and Google has issued manual actions against sites that scaled generated content.

Is a local agency better than a national one?

Not inherently, though proximity helps in two ways that are easy to underrate. The decision factors for a regional category (a swim school, a law firm, a contract manufacturer in West Michigan) are often things only a local person would think to check, and the sources the engines cite for regional prompts are often local directories, chambers and news outlets that a local agency already knows how to reach. A national agency with a strong method will do fine on either; what matters is whether the people doing the work can get subject-matter answers from you quickly, and that is easier when you can meet. We work in person in West Michigan and remotely elsewhere, and the method is the same in both cases.

How much should the first engagement cost?

Less than you fear and more than a tool subscription. The sensible first step is a fixed-fee audit that ends in a ranked plan, and the what-to-expect guide describes the components a fair one should include. Our audit is quoted after a short scoping call and is credited against the first month of a program, which is a structure worth asking any vendor for, because it means the audit has to be good enough that you want to continue. There is no independent survey of AEO agency pricing as of October 2026 that we have been able to verify, so treat any "industry average" quoted to you as the vendor's own number.

Which warning signs should end the conversation?

• A guaranteed position in ChatGPT or any engine, since the engines do not produce stable positions.
• A report built on a single run per prompt, or a vendor who cannot say how many runs sit behind a figure.
• Recommendations that could have been written without looking at your site, such as FAQ schema for everything; Ahrefs' test on 1,885 pages found FAQ and HowTo schema made no measurable difference on AI Mode or ChatGPT.
• Content volume as the main deliverable, especially generated pages at scale.
• Paid placements or review schemes described as earned mentions.
• No fact-check log, which means nobody has asked the engines what they currently get wrong about you.
• A twelve-month commitment before any audit has been done.

None of these makes a vendor dishonest, and some good agencies have one or two of them for reasons they can explain. Two or more together usually mean the work is being done by software and resold, and you can buy the software yourself for the price on its pricing page. If you want to compare that path, the agency versus tool guide lays out when a subscription is enough and when it is not, and the question list gives you fifteen things to ask on the call.

Frequently asked questions