Buyers guide

Every agency now claims AI recommends their clients. Here's how to check.

AI search claims are the easiest thing in this industry to fake, because almost nobody tests them. Here are five checks you can run yourself in an afternoon, plus the evidence standard to demand before you sign anything.

By Brandon Kelly · Updated July 2026 · 8 min read

To verify an SEO company's AI-search claims, ask for dated, reproducible citation examples (exact prompt, engine, and date), a written methodology, and proof that AI engines can find the agency itself. Then re-run their example prompts in ChatGPT and Perplexity yourself. If they can't show evidence you can reproduce, treat the claim as marketing.

Why "AI recommends our clients" is the easiest claim in SEO to fake

In the last few years, nearly every agency pitch grew an AI line. We get clients cited by ChatGPT. We rank you in AI Overviews. We do GEO. The claims multiplied much faster than the evidence, and there is a structural reason: AI search has no public scoreboard. There is no Search Console for chat engines. Answers shift from session to session. A screenshot proves very little on its own, because anyone can cherry-pick one lucky run, or type a prompt so narrow that only one business on earth could come back.

Full disclosure before we go further: this page lives on an SEO agency's site, and we include AI search optimization in every plan we sell. That is exactly why we wrote the checklist we would want used against us. Every test below is one we are prepared to pass on request, and anyone selling this work should be able to say the same.

Check 1: demand dated, reproducible citation examples

When an agency says its clients get recommended by AI, ask for the receipts, and be specific about what a receipt is:

  • The exact prompt, word for word. Not "we asked about plumbers." The actual sentence typed into the engine.
  • The engine and mode. ChatGPT with search on, Perplexity, Google AI Overviews. They behave differently, and the example should say which one it came from.
  • The date. An undated screenshot is a decoration, not evidence.
  • What actually appeared. Was the client named and cited with a link to their site, or did a category answer merely gesture at businesses like theirs?

Then run the prompt yourself, more than once. This is the step almost no buyer takes, and it costs five minutes. Expect variance: AI answers genuinely change between runs, so a fair test is three to five attempts spread across a day or two, not a single roll of the dice. A real citation shows up in a meaningful share of runs. A fake one was never going to survive your first attempt.

Watch for prompts engineered to be unloseable. "Best family-owned pool cage rescreening company near Vanderbilt Beach that answers on Sundays" is not a citation win; it is a question with one possible answer. The prompts that matter are the ones a real customer would type: who is a good plumber in Naples, a dentist for kids near Marco Island, a seawall contractor in Cape Coral.

Check 2: ask for the methodology in writing

Real AI-search work is boring and describable. If an agency cannot explain it in plain sentences, it probably is not doing it. As of mid-2026 the engines are not mysterious: ChatGPT's search leans on Bing's index, Perplexity retrieves live pages and shows numbered citations, and Google's AI Overviews draw from the normal Google index. Any methodology that does not route through crawlable, quotable pages is a story, not a method.

A credible written answer includes things like: unambiguous entity data, meaning the same business name, address, and phone everywhere the business is mentioned; pages structured so a direct answer comes before the detail; schema markup that tells machines exactly what the business is; review and citation signals that corroborate what the site claims; and crawler access so the engines can read all of it. Ours is published in the open on our AI Search Optimization service page, which is where any agency's methodology should live: in public, where you can compare the promise to the practice.

Answers that should end the meeting: "proprietary AI technology," "we submit your business to the AI engines" (that is not a thing), or any pivot back to a traffic chart when you asked about citations.

Check 3: test the agency's own AI visibility

An agency selling AI citability should be at least somewhat AI-citable. Ask ChatGPT and Perplexity about the firm by name: what does this company do, where is it, who does it work with? If the engines come back empty or wrong, the agency has not done for itself what it wants to charge you for.

Be fair about scale. A young or small firm will not dominate every category query, and that alone is not disqualifying; we are a young firm ourselves and say so plainly on this site. What you are checking for is a floor: a site the engines can retrieve, an accurate description when asked directly, citations that resolve to real pages the firm controls. An agency claiming years of AI expertise with zero machine-readable footprint is telling you something important.

Check 4: the two-minute crawler check

Before any of the subtler tests, look at the agency's own robots.txt file, because some firms selling AI optimization block the very crawlers they claim to court. We wrote a separate step-by-step for that test in how to check if an agency blocks AI crawlers, and a companion guide for testing your own site the same way. For the record: naplesseo.com welcomes GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, and Google-Extended, and publishes llms.txt and llms-full.txt files, verified July 2026. Hold anyone you interview to the same standard, us included.

Check 5: ask how they'll measure it after you sign

A claim you can verify once is good. A number you will see every month is better. Ask what the monthly report will actually contain for AI visibility: which prompts are tracked, on which engines, how often they are re-run, and how run-to-run variance is handled. The honest shape is appearance rates over time on a fixed prompt set, not a single triumphant screenshot per quarter. Ask, bluntly, whether flat months get reported. Then watch how they answer.

We keep a full guide to tracking AI visibility if you want to see what disciplined measurement looks like, and a companion buyers-guide page on reporting red flags for spotting the vanity-metric version before it costs you a year of fees.

The evidence standard to demand: dated, public, baseline first

Here is the bar we think every buyer should hold every agency to, including us: case studies published from day one, starting with the baseline, before there is anything to brag about. Dated entries. Flat months included. No invented numbers, ever, and no promised rankings, because nobody can honestly promise them.

We hold ourselves to that bar in public. We have exactly two clients, and both case studies went live at signing. SWFL Media Blasters signed June 29, 2026, and their entry documents the starting point and the build: a 47-page site with a 25-post blog and an optimized Google Business Profile. Pediatric Dentistry of Florida signed in July 2026, and their study opened the same way, baseline first. You can read both on our results page, and the SWFL Media Blasters study is a concrete example of what a day-one entry looks like.

Why does this matter for AI claims specifically? Because dated public records are the one format that cannot be retrofitted. An agency that claims three years of AI-citation wins should be able to produce three years of dated examples. If the evidence only exists as of the sales call, so does the expertise.

A test script you can run this afternoon

Pick one of the agency's named clients, say a roofer in Bonita Springs or a med spa in Naples, and work through this:

  1. Ask Perplexity a customer-shaped question about that client's trade and city. Check the numbered citations for the client's actual website.
  2. Ask ChatGPT, with search enabled, the same question. ChatGPT leans on Bing's index, so a business invisible to Bing rarely surfaces here, which is itself diagnostic.
  3. Run the same query on Google and read the AI Overview if one appears.
  4. Ask each engine about the client by name: what do they do, where, since when? Score the answers for accuracy, not just presence.
  5. Repeat the promising prompts two or three times across a day or two. You are looking for a business that keeps appearing, not one that appeared once.
  6. Bring your results to the agency and ask them to explain any citation you found: which page earned it, and why. Practitioners point at specific pages. Storytellers point at "the algorithm."

Total time: about an hour, most of it waiting on the engines. It is the cheapest due diligence you will ever do on a purchase that runs a few hundred to a few thousand dollars per month across this industry.

Where this fits in your vetting

AI citability is one station in a longer line. Contract terms, ownership on exit, and reporting honesty deserve the same scrutiny, and our umbrella guide on how to choose an SEO company walks the whole process in order. Run the checks on this page against anyone you are considering, including us. We built them to be passed, and the agencies that cannot pass them are the reason this page exists.

Frequently asked questions

Ask for the exact prompt, the engine, and the date for each claimed citation, then re-run the prompt yourself in that engine several times. A real citation reappears in a meaningful share of runs and links to pages the agency can explain. If the agency cannot produce dated, reproducible examples, or the prompt is written so narrowly that only one business could match it, treat the claim as marketing rather than evidence.
Variance is normal, so test in batches instead of single runs. Run the same prompt several times across a couple of days and note how often the business appears and whether its website is actually cited. Consistent presence across runs and engines is the signal; one lucky screenshot is not. That is why dated examples plus your own re-runs beat any sales deck.
No. Nobody controls what ChatGPT, Perplexity, or Google AI Overviews say, and answer positions are not stable the way old ranking reports implied. An honest agency commits to process and transparency: crawlable pages, clean entity data, answer-shaped content, and reporting that includes flat months. We put that in writing ourselves and would not sign with anyone who guarantees outcomes they cannot control.
It can be workable if they are honest about it. As of mid-2026 the fundamentals overlap heavily: AI engines favor clear, unambiguous entities and quotable, well-structured pages, which good SEO already builds. What you should not accept is a new practice sold with old evidence, like traffic charts presented as proof of AI citations. Ask for their first dated AI test on a real client, even a modest one.
A fixed set of tracked prompts per engine, re-run on a schedule, with appearance rates shown over time instead of one-off screenshots, and calls and forms reported next to visibility so you can see whether any of it matters. Baselines come first, and months with no movement still get reported. If a report only ever contains good news, that tells you more about the report than about the results.
Ready to get found?

See where you stand in Google and AI search

We will run a free audit of your local rankings and your visibility across ChatGPT, Gemini, Perplexity, and Google AI Overviews, then show you the fastest path to more booked jobs.

Call (239) 747-0465Free audit