We got mentioned once and then disappeared

What is actually happening

You asked ChatGPT a question your buyers ask. Your name came back. Somebody screenshotted it and put it in the team chat. A week later the same question returns three competitors and a blog roundup, and you are gone. Nothing was deleted. No penalty arrived. The answer simply changed.

An answer is a sample, not a ranking

A search engine returns a list that is broadly stable between refreshes. An AI assistant composes a fresh answer each time. It may run a live search, pick a handful of pages from the results, and write from those pages. Change the time of day, the phrasing, the user's location, or the model version, and the retrieved set changes with it. If your page sits at the edge of that set, you appear sometimes and not other times. One appearance tells you that you are reachable. It does not tell you how often you are reached.

Five apps answer the same question five ways

ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews read the web through different crawlers and different retrieval stacks. Each operator publishes its own crawler documentation, which is the cleanest primary evidence that these are separate pipelines rather than one shared index.

Last verified: 18 October 2026

Google also documents how its AI features use web content and what controls publishers have, in its AI features guidance. Four separate pipelines means four separate chances to be included or skipped. A win in one is not a win in all.

One mention often means one thin source path

When a brand appears once and then stops, the usual cause is that a single page carried it: one listicle, one forum thread, one directory row. If that page drops out of the retrieved set, the brand drops with it. Brands that appear steadily are usually named on several independent sites that the engines keep reaching for.

Asking about yourself proves nothing

The most common self inflicted error is typing "what do you think of [your brand]" and treating the glowing paragraph as evidence. Naming the business primes the engine. It will now search for that name and talk about it. Your buyer does not type your name. Your buyer types the problem. Every tracking question in AI Knows Us is asked without naming the business, for exactly this reason.

How to confirm it is happening to you

You need a count, not a screenshot. Follow these steps in order.

1. Write five to ten questions a buyer would type before they know your name. Example: "best payroll software for a 20 person agency in India". No brand names.
2. Pick the engines your buyers use. Start with ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews.
3. Decide a run size before you start. Twenty runs per question per engine is a workable floor. Write the number down so you cannot move it later.
4. Use a signed out, fresh session for every run. A logged in account carries memory and history, and memory will quietly re-prime the engine with your brand.
5. Record each run as a simple yes or no: did the answer name your business anywhere, including in citations.
6. Count the yes answers. Divide by total runs. That is your mention rate.
7. Put a 95 percent range around it. Use the Wilson score interval: (p + z²/2n ± z√(p(1-p)/n + z²/4n²)) / (1 + z²/n), with z = 1.96.
8. Repeat the identical set 30 days later. Compare the ranges, not the headline numbers.

Worked example you can repeat

One question, five engines, twenty runs each. That is 100 answers. Your brand is named in 7.

• Mention rate: 7 of 100 = 7 percent.
• Wilson 95 percent range: 3.4 percent to 13.7 percent.

Thirty days later you run the same 100 answers and get 9 mentions.

• Mention rate: 9 percent.
• Wilson 95 percent range: 4.8 percent to 16.2 percent.

The headline number went up by two points. The ranges overlap heavily, so the honest reading is: nothing moved. This is the rule AI Knows Us applies to its own reports. Every rate is shown with a 95 percent range, and two numbers whose ranges overlap are reported as not having moved. The same arithmetic also explains your original problem. At a true rate near 7 percent, a single lucky hit in your first five tries is ordinary, not a breakthrough.

What to change, in order

Order matters. Doing step four before step one wastes money.

1. Fix reachability first. Check robots.txt and any firewall or bot rules against the crawler names in the table above. If GPTBot or PerplexityBot is blocked, no amount of content work will register in those apps. This is a thirty minute job and it is the cheapest thing on this list.
2. Make one page per buyer question. Engines retrieve passages, not brands. A page that answers one question directly, with the answer in the first hundred words, is far easier to quote than a capabilities page.
3. Put checkable specifics on those pages. Named lists, dated prices with tax treatment, counts, per unit tables. Vague marketing copy gives a model nothing to lift.
4. Get named on sites you do not own. This is usually the step that moves a stuck rate, because a single owned page is a single retrieval path. Directories, review sites, trade bodies, supplier lists and roundups all count. See #.
5. Keep your existing search work running. Several of these engines run live web searches, so classic ranking and indexing still feed them. Read # before you move budget away from SEO.
6. Re-measure on a fixed clock. Same questions, same run size, same engines. Changing the question set between runs makes the comparison meaningless.

How long before it moves

Honest answer: we do not publish a guaranteed timeline, because the time depends on how fast each engine recrawls, how fast third party sites publish, and how competitive the question is. What we can be concrete about is the measurement clock. The free scan in AI Knows Us repeats every 30 days, so 30 days is the natural comparison window. Anything shorter and the ranges will almost always overlap, which means you will learn nothing.

Two practical expectations. First, reachability fixes can show up in the next crawl cycle, because they remove a hard block. Second, source building is slower, because it depends on other people publishing. Judge it over several 30 day cycles, not one.

Where AI Knows Us fits

AI Knows Us measures across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. It asks every tracking question without naming the business, prints a 95 percent range with every rate, and treats overlapping ranges as no change. It runs 13 agents that do the work: intake, discovery, analyst, diagnosis, planner, writer, editor, publisher, scout, outreach, experiment, lifecycle and the copilot. That means it does the fixing as well as the measuring: it writes the pages, publishes them, and asks other sites to name the business. It is made by Clyra Labs in India. Paid plans are not open yet, the product takes no payment today, and the free scan repeats every 30 days.

Other tools in this category

Several products track brand visibility in AI answers. Check each one's own site for current features and pricing, because we do not hold dated pricing for them and will not guess.

• Profound
• Peec AI
• Otterly.AI
• Scrunch AI

Who should not choose AI Knows Us

• Teams that need a signed enterprise contract and a paid SLA today. Paid plans are not open yet.
• Teams that want a dashboard feed only, with their own agency writing and placing everything. The agent set is built to do the work, not just report on it.
• Teams that want a single confident percentage with no range attached. Every number here comes with a range, and some readings will honestly say "no change".
• Teams whose buyers never use AI assistants. Classic search work will pay back faster.

FAQs