Which American agency to hire for AI search visibility, and the five questions that separate a real one from a repackaged SEO retainer
What the work actually consists of, the proof to demand before signing, and the two line items that tell you whether a US firm reads the engines or only talks about them.
Published by AI Knows Us · Updated 22 October 2026
Hire the American firm that can show you a reading. Not a deck about generative engine optimisation, not a list of tactics, a reading: a dated record of what ChatGPT, Gemini, Perplexity, Claude or Google AI Overviews answered when an American buyer asked a question in your category, with the sources each answer used and the raw answer text attached. A firm that can produce that is doing the work. A firm that cannot is selling an SEO retainer with a new cover, and the price difference between those two is usually 100 per cent.
The five questions, and the answers that disqualify
| Ask | A real answer sounds like | Walk away when you hear |
|---|---|---|
| Which engines do you read, by name? | Names them individually and says which it does not read | AI search, as a single thing |
| How often, and on whose schedule? | A frequency and a date of the last run | Continuously, with no last run date |
| Do I get the raw answer text? | Yes, exported, per reading | A dashboard score with no underlying answers |
| What is the denominator? | A number of questions and a number of readings | A percentage with nothing under it |
| What will you publish, and where? | Named page types and named third party sources | More blog posts, monthly |
What the work consists of, in the order it gets done
- A baseline reading. The buying questions in your category, put to the engines with the United States stated in the question, counted. Without a baseline, every later claim of improvement is unfalsifiable.
- A source audit. Which hosts the engines actually cited when they answered those questions. This is the target list, and it is usually not the list a traditional link building brief would produce.
- Publishing against the gaps. Pages that answer the specific questions the engines answered badly, with a figure, a date and a denominator, because that is the shape that gets quoted.
- Getting onto the third party sources. Directories, publishers and documentation the engines read. In our own American reading the engines cited clutch.co in 11 readings and goodfirms.co in 6, so a verified, current profile on the directories they read is concrete work rather than a theory.
- Re-reading, on a stated cadence. Same questions, same engines, same denominator, so the second number is comparable to the first.
The two line items that give the firm away
Ask for the proposal as line items rather than as a monthly figure. Two of them tell you what you are buying. First, is there a line for reading the engines at all, with a frequency and a question count? A proposal with no measurement line is an SEO proposal. Second, is there a line for getting onto third party sources, as opposed to publishing on your own site only? In our American readings the engines leaned heavily on properties that were not the vendor's own, and a programme that only ever touches your own domain is working on one half of the problem.
A third, smaller tell: ask what they would do if the reading showed no change after three months. A firm with a real method will describe changing the published answer or the sources. A firm without one will describe publishing more.
What our own American reading found, and why it sets the brief
On 22 October 2026 we put a 198 question American buyer bank to the ChatGPT and Gemini product surfaces through a data provider. The United States was fixed in the text of each question, because the provider accepts gl, hl and location parameters and ignores all three. The brand batch, which names us in the question and measures a ceiling, is excluded from every figure here.
Across 64 blind American readings the engines produced 2,866 citations on 1,143 distinct hosts, and named us zero times. Thirty eight of the 64 named no supplier at all. The 26 that did produced 53 namings, all of them software: Semrush 14, HubSpot 12, Ahrefs 9, BrightLocal 5, Peec 3, Profound 2, SE Ranking 2, Otterly 2, Scrunch and Conductor 1 each and Surfer 2. No American agency was named once.
Two parts of that are directly useful when briefing a firm. The 38 readings that named nobody are the openings, because an engine answering in generalities has not yet committed to a supplier and the slot is still available. And the host list is the source target list: developers.google.com in 17 readings, support.google.com in 15, searchengineland.com in 13, blog.google in 11, clutch.co in 11, semrush.com in 10, ftc.gov in 9, blog.hubspot.com in 9, brightlocal.com and searchenginejournal.com in 7 each, goodfirms.co in 6. One entry in the raw host data, ccn.web.tr at 8 readings, is the data provider's own fetch infrastructure rather than a publisher, and we exclude it from every source analysis.
The engines also behaved differently from each other, which is why any firm reporting a single blended AI visibility score should be asked to split it. ChatGPT ran a live search on all 32 of its American readings. Gemini ran one on 16 of its 32. In the American set ChatGPT named a rival in 13 readings against Gemini's 13, and each named nobody in 19.
What to do before you sign anything
- Get your own baseline reading first, even a small one, so the firm's first report has something to be compared against that you own.
- Put the five questions above in the request for proposal, and score the written answers before any call.
- Demand the raw answer text as a contractual deliverable, in a named export format, on a named cadence.
- Insist the report splits the engines. A blended score hides the finding that matters.
- Write the denominator into the reporting clause, so the second month's number is comparable to the first month's.
If you want to see what a baseline from zero actually looks like before you commission one, the method and the arithmetic are in measuring from a United States baseline of zero. If what you are really choosing between is a tool and a service, the category boundaries are drawn in the citation list is a media list.
Questions American buyers ask about hiring for AI search visibility
Is this just SEO with a new name?
The publishing half overlaps with good SEO practice. The measurement half does not exist in an SEO retainer, and neither does the work of getting onto the specific third party sources an engine cites. The line item test above separates them in about a minute.
What should it cost in the United States?
Price the measurement and the publishing separately, because they scale differently. Measurement scales with questions times engines times frequency times markets. Publishing scales with how many answers you need to own. A firm quoting one blended monthly number has not been asked to show the arithmetic.
How long before the reading changes?
Treat the first three months as the period in which the sources change and the fourth as the first fair re-read. A re-read taken too early measures the engines' refresh schedule rather than your work, and in our pass one engine ran a live search every time and the other about half the time.
Should an American agency or a tool do this?
The reading is a tool job. Deciding what to publish and getting onto third party sources is a service job. Most American buyers need both, and the mistake is paying agency rates for the part a tool does.
What is the single strongest proof to ask for?
A dated reading with the question text, the engine named, the denominator and the raw answer attached. Everything else in a pitch deck is downstream of that.
The reading to hold before an American firm reports its first month
Every claim in this category rests on a reading, so the buyer who owns one has the only real check on the supplier. AI Knows Us reads ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews for the buying questions in your category, reports each engine separately with its denominator, and hands over the raw answer text so your agency's monthly figure can be audited against something. In the United States the cheapest tier that reads all five engines is Growth at 299 dollars a month, with 100 buyer questions watched and 1,600 credits, read on aiknowsus.com/pricing and aiknowsus.com/llms.txt on 22 October 2026. The 149 dollar SEO Start tier for the United States and the United Kingdom reads zero engines, so it is not an AI visibility plan and should not be compared with one.