SIGNALS
Off-site signals

How do you get recommended by AI if you can't name your customers?

The short version

Getting recommended by AI without naming your customers is normal rather than a handicap, because the customer name is not what the engine reads. Most of what an answer engine quotes about a company sits on domains the company does not own, and an NDA does not reach those at all. What an NDA does block is the named logo wall and the named case study, and both are replaceable: a sector descriptor, a real number, the method behind it and a stated caveat carry the same weight with a reader and more weight with a verifier. The one thing never to do is manufacture the missing independence, which is now against the published usage policy of at least one major model provider. Publish the anonymised version, get the specifics quoted somewhere you do not own, and leave the logos to the clients who sign off.

How do you get recommended by AI if you can't name your customers?

Getting recommended by AI without naming your customers works because the customer name was never the signal. An engine assembling a shortlist reads text about you, wherever that text sits, and weighs what it can attribute. A logo on your homepage is an image with no claim attached to it, and a named case study is one page on a domain you control.

Where the weight actually sits is measurable. Muck Rack analysed more than 25,000,000 links cited by ChatGPT, Claude and Gemini and found about 84% of citations came from earned sources, counting journalism, academic work, government and non-profit sites, encyclopedic references, social platforms and third-party corporate content, while owned and paid content together accounted for about 16%, in "What is AI Reading?" (May 2026, read 19 November 2026).

An NDA governs what you may say about a specific relationship. It has no purchase on whether a trade publication quotes your head of engineering, whether your category's comparison pages list you, or whether somebody answers a question in a forum and names you. That is the larger share of the retrieval surface and it is untouched by the restriction you are worried about.

So the problem is narrower than it first looks. You need proof a stranger can check, in words that survive being lifted out of context, published somewhere an engine already reads. None of those three requirements contains a customer name.

Do AI engines read your case studies and customer logos at all?

Case studies get read as ordinary pages and customer logos mostly do not get read at all, because a logo wall is a row of images with no sentence an engine can quote. A case study with a number, a method and a date is a different object: it is a page carrying a checkable claim, which is the form engines reward.

What survives controls here is vocabulary rather than branding. The largest published analysis of citation behaviour reports a standardised coefficient of +0.37 for content alignment, meaning overlap between the page's language and the way buyers phrase the question, at 95% confidence bounds of +0.33 to +0.41 and q below 1e-73, across about 2 million citation observations and 10,000 cited pages on ChatGPT, Claude, Google AI and Gemini, according to Discovered Labs, "What actually drives AI citations" (2026, read 19 November 2026). Discovered Labs sells AEO services, so the choice of what to measure is not disinterested, and the method is unusually explicit about its controls.

Two practical consequences follow. A case study titled with the client's name buys nothing, because nobody is asking a question in those words, and a case study titled with the problem and the measured result matches a question somebody types. And the anonymised version loses almost nothing that an engine was using, since what it quotes is the sentence with the number in it.

Paid placement does not rescue the logo wall either. The same Muck Rack study found that paid and advertorial content accounted for about 0.3% of all citations, which makes bought credibility the least efficient route available.

What can you publish instead of a named case study?

Instead of a named case study, publish the same evidence with the identity removed and the method left in. A sector descriptor, the size of the business, the problem in the buyer's own words, what was changed, the number that moved, the period it moved over, and a sentence saying what the number does not prove. Readers accept that shape, and it is more useful to a verifier than a logo.

Our own results page is built that way, which is the only reason we can point to it here. It describes two early clients as a tech and QA founder and a video producer, reports one audit score moving from 17 to 68 and the other more than doubling, and states that the second is given as a multiple rather than a pair of numbers so the client stays genuinely unidentifiable, on our early results page. The same page says plainly that the audit score is an input metric describing how well a site is built for citation, while share of voice is the outcome, and that a strong score stacks the odds rather than guaranteeing a citation.

Where no client can be described at all, use yourself. Our teardown of our own page works through a score moving from 28 to 71 on the same framework, with the before and after visible, and it needed nobody's permission.

One rule holds across all of it. Keep the mapping from anonymised story to real client in a restricted internal document, and check that the identifying detail has not survived into a filename, an image or a stray industry specific. An anonymised study that identifies the client by accident is worse than no study, because it tells every future prospect what you will do with their name.

Why do the brands that get cited differ from the brands that get recommended?

Brands that get cited and brands that get recommended come apart because citation is a retrieval decision and recommendation is a judgement about standing. A page of yours can be the source an answer is built from while the answer names somebody else, and the measurement for that is now published.

Lily Ray checked 100 B2B "best [category] software" queries in Google AI Overviews on three dates between April and June 2026 and found that of 323 cited self-promotional listicles, 224, or 69%, belonged to brands that were left out of the recommendation itself, with 74 of the 100 prompts containing at least one such case, in "Why Calling Yourself the Best Could Be Helping Your Competitors Win in AI Search" (17 June 2026, read 19 November 2026). She reports that the brands that did get recommended generally had more referring domains and more brand mentions than the cited-but-unrecommended ones, and is explicit that the mechanism cannot be confirmed from outside Google.

Read that as a constraint on what your own pages can do rather than as a reason to stop publishing them. The anonymised case study earns the citation and the trust of a human reader. Being named in the recommendation tracks how widely you are described elsewhere, and that is the lane where the NDA costs you nothing.

Our write-up on how to get into the best-of lists AI engines quote covers the pitching side of that, including how to find which lists are actually being cited for your category before spending time on any of them.

Where does third-party proof come from when clients can't be named?

Third-party proof without client names comes from five places, and four of them need no customer to agree to anything. Reviews written by individuals under their own names rather than their employers', trade and analyst coverage of the problem you solve, original research you publish and others cite, conference and podcast appearances with a transcript, and answers you give in the communities where your buyers compare options.

Proof you might want Available when the client cannot be named Reaches an engine as
Named logo wall No, needs sign-off per logo Images with no quotable claim
Named case study No, needs sign-off and legal review A page on your own domain
Anonymised case study with a number Yes, usually with a quick client read A quotable sentence on your own domain
Individual review on a platform Yes, the reviewer speaks for themselves Third-party page, often cited directly
Trade or analyst coverage Yes, about the problem rather than the account Earned media, the largest cited category
Original research you publish Yes, uses your data rather than a client story Cited by others, then cited by engines
Community answers under a real name Yes, with the affiliation stated Forum and social sources engines retrieve

Review platforms deserve a specific note, because the NDA objection usually does not apply to them. The person writing the review is speaking for themselves, and most platforms allow a role and industry in place of a company name. Our answer on whether G2 and Capterra reviews affect what AI recommends covers how much that is worth.

If you want to know which of these your category's engines are actually reading before investing in any of them, a free visibility assessment samples your buying questions across the engines and reports which sources the answers were built from. The list is usually shorter and stranger than the plan anybody would have written.

What should you ask a client for when the NDA blocks a case study?

Ask the client for the smallest permission that still produces something publishable, and ask for it as a ladder rather than a yes or no. Full name and logo. Name in the text without the logo. Sector, size and result with no name. Result only, with the sector blurred one level. Most clients who refuse the first will approve the third in a single email, because nothing in it identifies them.

Timing helps more than wording. Ask while the result is fresh and the sponsor is pleased, include the exact draft paragraph you want approved, and tell them which number you intend to publish and which you will leave out. A request that reads like a form takes weeks; a request that reads like a two-sentence proof takes a day.

Where the answer is still no, do not manufacture the independence you are missing. Anthropic's usage policy, effective 12 November 2026, prohibits manipulating the sources from which search engines or AI systems draw answers by seeding them with content that misrepresents its origin, authorship or independence, in its published usage policy (read 19 November 2026). A fabricated review or an unlabelled site posing as a neutral comparison is squarely inside that description, and it is the one failure here that cannot be undone quietly.

The honest version works better than it reads. A line saying details are withheld at the client's request, next to a number you can stand behind and a method anybody can repeat, is a statement about how you treat confidential work. Buyers in NDA-heavy categories read it as the reason to call you.

What else do people ask about proof when you can't name customers?

Teams that cannot name customers ask these five questions most often about proof and AI recommendations.

Does an anonymised case study still help you get cited by AI?

Yes, because what gets quoted is the sentence carrying the number and the method, not the client's name. An anonymised study keeps both. What it loses is some persuasive force with a human reader, and the fix for that is specificity about everything except identity: the sector, the size, the starting position, the period and what the number does not prove.

Can you use a client's logo if they signed an NDA?

An NDA is about confidentiality and is not a grant of publicity rights, so the absence of a ban is not permission. Logo use needs its own sign-off, usually from marketing or legal rather than your project sponsor. Ask for it separately from the case study, because the two approvals move at different speeds and bundling them slows the faster one down.

What should you do if you have no customers to reference yet?

Publish on your own work instead. A teardown of your own site or product, original research from data you already hold, and a written method anybody can repeat are all available on day one, and all three produce the quotable, checkable sentences engines use. We did exactly that with a public teardown of our own page before we had results to show.

Do review platforms work when clients cannot be named publicly?

Usually yes, because the reviewer is speaking for themselves rather than for their employer, and most platforms accept a job title and industry in place of a company name. The practical blocker is procurement policy at large enterprises rather than the NDA, so ask the individual rather than the account team, and never offer anything in exchange for a review.

Is it worth paying for placement if you cannot show named proof?

Rarely, on the published evidence. Paid and advertorial content accounted for about 0.3% of citations in Muck Rack's May 2026 study of more than 25 million cited links, so bought placement is the least efficient way to buy presence. The same money spent on original research or on being quoted in trade coverage reaches the categories engines actually cite.

Related guides

The Assessment

See which sources the answers about your category are built from.

A free visibility assessment samples your buying questions across ChatGPT, Perplexity, Google AI Overviews and Claude, and reads your public pages the way a crawler sees them.

Request a free visibility assessment →
SIGNALS · A BlackSig Systems company