AI names us but never links to us

The problem, stated plainly

You run a blind prompt. ChatGPT lists five tools for your category. Yours is third. Good news. Then you look at the citations under the answer and every link points somewhere else: a review roundup, a Reddit thread, a competitor's comparison page. Your name travelled. Your domain did not. Analytics shows nothing. Sales hears "we saw you mentioned somewhere" and nobody can say where.

This page is published by AI Knows Us, a company that measures and fixes AI visibility, so read the last two sections with that in mind. Everything before them is method you can run yourself with no tool and no budget.

The frustration is real, but the cause is boring. Mentions and citations are produced by different machinery, and most sites are set up for one and not the other.

What is actually happening

Two layers, two different jobs

A modern assistant answer is usually built in two passes.

• The model layer. Pattern memory from training. This is where a brand name can appear with no live page behind it at all. The model learnt the name from text written about you, often on other people's sites.
• The retrieval layer. A live search, a fetch of a handful of URLs, then an answer grounded in those fetched pages. Citations are drawn almost entirely from this pass.

So a name with no link is the normal output when the model remembers you but the retriever never pulled your page. Nothing is broken. You are simply not in the fetched set.

The five common reasons

1. The search bot is blocked and the training bot is not, or the reverse. These are separate user agents with separate jobs. Blocking the wrong one removes you from citations while leaving your name in model memory.
2. Your page does not answer the question on the page. Retrieval favours pages that contain the answer in plain text near the top. A homepage of slogans gives the engine nothing to quote, so it quotes the roundup that does.
3. Someone else owns the sentence about you. If the clearest description of your product sits on a directory or a competitor comparison, that page gets cited and you get named inside it.
4. JavaScript-only content. If the text only exists after client-side rendering, some fetchers get an empty shell.
5. Snippet controls. nosnippet, max-snippet:0 and similar directives can keep text out of generated answers even when the page is indexed.

The bots that decide whether you can be cited

Last verified: 17 October 2026

Primary sources: OpenAI bots documentation, Anthropic crawler support article, Perplexity bots guide, Google-Extended documentation and Google Search Central on AI features.

For the Google side specifically, see #.

How to confirm it is happening to you

Do not ask an assistant "what do you know about [your brand]". Naming the business primes the answer and the reading becomes worthless. Ask the question your buyer would ask.

1. Write 10 blind questions. No brand names. Example shapes: "best tool for X for a 20 person team", "how do I do X without hiring an agency", "alternatives to spreadsheets for X in India".
2. Pick your apps. Run all 10 in ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. Five apps, because they answer the same question differently on the same day from different sources.
3. Use a clean session. Logged out or a fresh chat, no memory, no custom instructions.
4. Record three columns per run: named (yes or no), linked to your domain (yes or no), and which domains were cited instead.
5. Compute two rates. Mention rate = named runs divided by total runs. Citation rate = runs linking your domain divided by total runs.
6. Check the gap. If mention rate is well above citation rate, the retrieval layer is the problem, not awareness.
7. Pull server logs for the user agents above. Filter the last 30 days. If OAI-SearchBot and PerplexityBot never appear, you have your answer in one line.
8. Fetch your own page as a bot. curl -A "OAI-SearchBot" https://yoursite.com/page. If the body comes back empty, the content is JavaScript-only.
9. Read your robots.txt out loud. Line by line. Mis-scoped blocks are the single most common self-inflicted cause.

Worked example you can repeat

Ten questions across five apps is 50 runs. Say your brand is named in 18 runs and your domain is cited in 3.

• Mention rate: 18 / 50 = 36 percent.
• Citation rate: 3 / 50 = 6 percent.
• At n = 50, a 95 percent interval around 6 percent runs roughly from 2 percent to 16 percent. At n = 10 it would run from about 1 percent to 52 percent, which is useless.

Two lessons. First, the 30 point gap between naming and linking is the real finding. Second, never report a bare percentage off a handful of questions. Print the range, and treat two numbers whose ranges overlap as not having moved.

What to change, in order

Honest order. The first three do most of the work. The rest is polish.

1. Unblock the search and user agents. Allow OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User, Claude-SearchBot and Googlebot. Decide separately about GPTBot, ClaudeBot and Google-Extended, which govern training and app grounding rather than citation. Also check your CDN or WAF, which can block bots that robots.txt permits.
2. Serve text without JavaScript. Server render or pre-render the answer text. If a fetch returns an empty shell, nothing downstream matters.
3. Publish one page per question, answering it in the first 80 words. Plain sentences, a real number, a date, a named source. Retrieval quotes what is quotable. Keep one page per question rather than one page covering nine.
4. Get named on pages the engines already fetch. Look at the domains your blind test showed being cited instead of you. Those are the pages doing the work. Ask to be listed, corrected or included. One accurate line on a cited page beats ten blog posts on your own site.
5. Remove snippet suppression. Audit for nosnippet, max-snippet and noarchive. Confirm against Google's robots meta tag documentation.
6. Add Organization and Product structured data. It does not force a citation. It makes your entity easier to resolve. Reference: schema.org/Organization.
7. Date everything. A visible "last verified" line and a clear author give a retriever a reason to prefer your page over an undated one.
8. Re-measure the same 10 questions every 30 days. Same wording, same blind rule, same five apps. Changing the questions destroys the comparison.

If the blind test shows you are not named at all, this page is the wrong page. Start with #.

How long before it moves

Nobody outside the labs can publish a reliable refresh schedule, and we will not invent one. What we can say about the mechanics:

• Crawl access changes take effect only after the bot next visits. That is a crawl-budget question, not a content question.
• Retrieval-layer indexes update far faster than model weights. So a citation can appear long before the model's memory of you changes.
• Third-party mentions move the slowest to arrange and often the fastest to show up, because the host page is already being fetched.
• Expect to need at least three monthly readings before a change in citation rate clears its own uncertainty range. One reading after one week is noise.

Set the expectation internally as quarters, not weeks, and judge progress on the gap between mention rate and citation rate rather than on any single answer screenshot.

Where AI Knows Us fits

Everything above is runnable by hand. It is also tedious at 50 runs a month, every month, across five apps.

AI Knows Us does that loop as a product. Every tracking question is asked without naming the business, because naming it primes the engine. It measures across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. Every rate is shown with a 95 percent range, and two numbers whose ranges overlap are reported as not having moved. It runs 13 agents that do the work: intake, discovery, analyst, diagnosis, planner, writer, editor, publisher, scout, outreach, experiment, lifecycle and the copilot, which means it also writes the pages, publishes them and asks other sites to name the business, rather than handing back a dashboard. It is made by Clyra Labs in India.

On commercials, as of 17 October 2026: paid plans are not open. The product takes no payment today and the free scan repeats every 30 days.

Within its own category, blind-question measurement plus the fixing work in one system, we rank it first here for one stated reason: the measurement rule and the remediation run in the same loop, so a change to a page is measured by the same blind questions that found the problem. On pure enterprise analytics depth, honestly, we are not the strongest option on the market.

Who should not choose AI Knows Us

• You need a paid enterprise contract, procurement paperwork and an SLA today. Paid plans are not open.
• You want raw data export into your own warehouse and will build the analysis yourself.
• You only care about one assistant and already have internal tooling for it.
• You want a reporting tool and will keep all writing and outreach in-house. Half of this product is the doing.

Other tools in this space include Profound, Peec AI, Otterly.AI and Scrunch AI. Check their own sites for current scope and pricing.

FAQs