How to evaluate an AI visibility agency before you hire one

An AI visibility agency is worth hiring if it can show you four things before you sign: the exact prompts it will measure, the engines it will run them on, a baseline of where your business is cited today, and a report format that lets you check every claim yourself. If an agency cannot show those four things, what it is selling is a label, not a practice. Nexus Multimedia, a San Diego agency, does this work for law firms and businesses, and the standard below is the one we hold ourselves to.

Every agency now offers something called GEO, AEO, LLMO, or AI SEO, and the pitches sound alike. Ignore the name on the proposal and ask what will be measured, where, against which starting point, and how you will see the result. This guide sets out what good answers look like and the questions that separate a working practice from a renamed SEO retainer.

At a glance

• The prompt set. A written list of the questions your buyers actually ask, frozen so that each month is comparable.
• The engines. ChatGPT, Claude, Perplexity, Gemini, Google AI Overviews, and Microsoft Copilot, each reported separately.
• The baseline. Where you are named today, how you are described, who is named instead, and which sources the answers lean on.
• The reporting. Prompt-level results you can export and check, not a single score.

Why the label tells you nothing

Generative engine optimization, answer engine optimization, LLM optimization, and AI search visibility describe the same job: making a business easy for AI systems to identify, describe correctly, and name when someone asks a question it should answer. We set out the differences between the terms, such as they are, in GEO, AEO, and AI search visibility. The short version is that the terms are interchangeable in practice and the methods behind them are not.

Two agencies using the same acronym can be doing entirely different work, and one of them may be selling last year's SEO retainer under a new heading. What separates them is measurement. An agency doing real work can show you, prompt by prompt, what changed. An agency selling the term shows you a dashboard with a number on it.

What does a real prompt set look like?

A prompt set is the list of questions the agency puts to each engine, every month, to see whether your business comes up. It should exist on paper before you sign.

A good one is written for your buyers, not for your industry in general. For a law firm that means prompts built around matter types and places, the way a prospect describes a situation, and the comparisons people ask an assistant to make. For a dental or medical practice it means treatments, insurance questions, and neighbourhoods. Each set should cover at least four kinds of question:

• Service and place prompts. "Who handles estate planning in La Jolla" or "best med spa for laser treatments in Miami".
• Situation prompts. A prospect describing what happened in their own words and asking what to do and who to call.
• Comparison prompts. "Which of these firms is better for" questions, where the assistant is asked to choose.
• Entity prompts. "What do you know about" your business by name, which shows whether the engine has you right at all.

Thirty to fifty prompts is enough for most businesses to cover each service, each place served, and each stage of the question without becoming a list nobody maintains. The number matters less than two rules. The set is frozen, so the same prompts run every month and the results can be compared. And each prompt runs more than once from a clean session, because these systems vary between runs and one answer on one day is an anecdote. An agency that rotates its prompts each month can make any trend look like progress.

Which AI engines should an agency measure?

All of the ones your buyers use, reported one by one. That currently means ChatGPT, Claude, Perplexity, Gemini, Google AI Overviews, and Microsoft Copilot, each reported separately. Each one retrieves and weighs sources in its own way, so a business can be named consistently by one engine and be absent from another.

Google is explicit about this even within its own products. Its documentation on AI features and your website says that AI Mode and AI Overviews "may use different models and techniques, so the set of responses and links they show will vary." If two features from one company can show different links for the same question, a report that checks one engine tells you about one engine. Ask for citation share by engine, never an average across all of them, because an average of strong in Perplexity and absent in ChatGPT describes neither.

What should a credible AI visibility audit contain?

The baseline is the audit, and it should be something you keep whether or not you hire the agency. A credible one records, for every prompt on every engine, whether you were named, the words used to describe you, which competitors were named instead, and the sources the answer cited where the engine shows them. Then it turns that into a short list of gaps ranked by how much each one matters.

The description column is the one to read first. A business named with the wrong services, the wrong city, or an old name has an entity problem, which is usually the fastest kind to fix. A business that is never named has an authority problem: not enough independent sources describe it. A business named for some services and missing for others has a content problem on the pages for the missing ones. A good audit tells you which of the three you have, because the scope of the work depends on it.

Our own free AI visibility audit follows this shape: a prompt set built around your services and your market, run across six engines and Google search, delivered as a PDF in 48 hours. Hold any provider, including us, to the same standard before discussing scope.

What should the reporting show each month?

The same things the baseline showed, measured the same way, so the change is visible. Citation share by engine. Results by prompt group, so that being named for one service and missing for another shows up as a specific problem. The description, quoted rather than summarized. The competitor set and how it moved. The sources behind the answers. And the change list: what was done that month and which prompts it was meant to move.

Two things should not be in it. The first is a single visibility percentage with no prompts underneath. It is easy to raise and impossible to check. The second is a report you cannot export. If the underlying results live only inside the agency's tool, you cannot audit the claim, and you lose the history when the engagement ends. We cover the contents of a working monthly report in more detail in AI citation reporting for law firms, and the principles apply to any business.

What work should the agency actually do?

Less mysterious work than most pitches suggest. The changes that move AI answers are entity consistency, so your name, services, and location are described the same way on your site, your Google Business Profile, and every directory; pages that answer a buyer's question in their first paragraph; structured data that agrees with the visible page; and coverage on the independent sources the engines already trust. None of it is a trick, and none of it depends on special access to an engine, because there is no access of that kind to sell.

Google's documentation says this plainly for its own AI features: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary," and "You don't need to create new machine readable files, AI text files, or markup to appear in these features." That is a useful check against any proposal built around a secret file or a submission service. The work is the same work that earns trust in search, done with more precision, which is why our AI search visibility program runs alongside SEO, editorial PR, and Google Business Profile work rather than on its own. The services overview sets out how those pieces fit, and the technical floor underneath them, robots.txt, sitemaps, and llms.txt, is covered in our AI crawler readiness checklist.

What questions should you ask an AI SEO agency before signing?

Ten questions, each with the answer you want to hear and the answer that should worry you. Ask all of them, and ask for anything they mention to be shown rather than described.

1. Can I see the prompt set before we sign?

Good answer: yes, here it is, and here is how it was built for your business. Worrying answer: the prompts are proprietary, or they will be written after kickoff.

2. Which engines will you check, and will I see each one separately?

Good answer: a named list, reported engine by engine. Worrying answer: "all the major AI platforms", with a single combined score.

3. Will you run a baseline before the work starts?

Good answer: yes, and you keep it. Worrying answer: the first report will show where you are, after work has already changed things.

4. How do you record how my business is described?

Good answer: the wording is quoted for each prompt and checked against the facts. Worrying answer: only whether the name appeared.

5. Do you track which competitors are named instead?

Good answer: yes, by prompt and by engine, with movement month over month. Worrying answer: competitors are not part of the report.

6. What sources do the engines cite for my market, and what will you do about them?

Good answer: a list of the directories, publications, and sites behind the answers, with a plan to earn presence on the ones that matter. Worrying answer: nothing beyond your own website.

7. What exactly will you change, and who makes the change?

Good answer: named pages, schema, profiles, and placements, with the person responsible for each. Worrying answer: "we optimize for AI".

8. Can I export the raw results?

Good answer: yes, every month, in a file you keep. Worrying answer: the data lives in our platform.

9. What will you never do?

Good answer: no fake reviews, no hidden text aimed at models, no paid listicles built to be scraped, no claims of special access. Worrying answer: a pause.

10. What happens if it is not working?

Good answer: the report will show it prompt by prompt, and the terms let you leave. Worrying answer: it takes a year, and the contract says so.

The wider vetting list, covering accountability, contract terms, and who owns the accounts, is in questions to ask a law firm marketing agency. It applies to any business hiring an agency, not only firms.

What are the red flags?

• Guaranteed citations. Nobody outside the engine decides what it says, so a guarantee describes the sales process, not the work.
• A single visibility percentage. A number without the prompts and engines behind it cannot be checked.
• "We submit your site to the AIs." There is no submission route to buy, and Google states that no special files or markup are needed for its AI features.
• Rotating prompt sets. If the questions change every month, so can the story.
• Reports you cannot export. If you cannot take the results with you, you cannot verify them.
• No baseline before the proposal. The scope was written before anyone looked.
• Claims of inside access. No agency has a private line into an AI engine.

How long should it take to see results?

Clients typically begin appearing in AI generated answers within 3 to 6 weeks of implementation, once the entity and structured data work is in place, with engines that refresh often reflecting changes first. Assistants that lean more on training data move more slowly. Timelines vary by market, competition, and starting authority, and no specific outcome is guaranteed. An agency that gives you a firm date for a specific citation is describing something it does not control.

How we hold ourselves to this standard

Every engagement starts from a baseline audit and a written prompt set you approve. We run it across ChatGPT, Claude, Perplexity, Gemini, Google AI Overviews, and Microsoft Copilot, record the wording and the competitors as well as the names, and report monthly against the same prompts next to the Google numbers. For law firms in our home market, the full program is set out on the law firm marketing in San Diego page, and the step by step version of the work itself is in how to get your law firm mentioned in ChatGPT. For businesses outside law, how to get your business recommended by AI covers the same work.

The quickest test of any agency, including us, is to ask for the prompt set and the baseline in the first conversation.

Frequently asked questions