AI keeps citing our competitor's blog about us

What is actually happening

You type your category into ChatGPT. The answer names you once, then footnotes a competitor's "X vs Y" post as the source for what you do. It feels like sabotage. It is usually something duller.

Modern assistants answer in two ways. Some answers come from the model's training. Many now come from a live retrieval step: the assistant runs a search, pulls a handful of pages, and writes a summary with citations attached. OpenAI described this retrieval behaviour when it announced ChatGPT search on 31 October 2024 (OpenAI, Introducing ChatGPT search). Google documents the same pattern for AI features in Search (Google Search Central, AI features and your website).

So the question is not "why does the model hate us". It is "which page did the retriever find, and why was that page easier to quote than ours".

Five ordinary reasons the competitor page wins

• It answers the comparison question literally. Their page is titled with the comparison. Your page is titled with your brand promise. The retriever matches the question, not the ambition.
• It contains you by name, in the same paragraph as the category. That co-occurrence is what makes a passage look relevant to a question about you.
• It is extractable. Short paragraphs, a table, a dated price, a list. Your page may have the same information spread across a hero image, a video and a sales call.
• It is repeated elsewhere. A directory, a newsletter and two forum threads echo their framing. Retrieval rewards the version that several sources agree on.
• Your pages may not be crawlable by the agents that matter. Each assistant ships its own user agents and its own controls. See the table below.

A related case is when the cited source is a forum thread rather than a competitor. The diagnosis overlaps but the fix differs: see /blog/ai-cites-a-reddit-thread-about-us-that-is-wrong.

Crawler controls, per assistant

Check these against your robots.txt and your CDN rules. Named agents change, so treat the official page as the authority on the day you read it.

How to confirm it is happening to you

Do not ask an assistant "what do you think of [your brand]". Naming yourself primes the answer. The engine will fetch your own site and hand back your own marketing. You will feel reassured and learn nothing.

The test, in order

1. Write three blind questions. Each one must describe the buying situation and never name your company. Example: "Which tools help a B2B SaaS team track how AI assistants describe their brand?"
2. Pick five surfaces. ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. They answer the same question differently on the same day because they read different sources.
3. Use a clean session for every question. New chat, memory and personalisation off where the setting exists, logged out where possible. Chat history changes answers.
4. Run each question twice, on two different days. Three questions, five surfaces, two days gives 30 answers. That is the smallest run worth reading.
5. Record four fields per answer: date, surface, whether you were named, and every cited domain in the order shown.
6. Tally by domain, not by feeling. Count how many of the 30 answers cite the competitor's blog, and how many cite any page on your own domain.
7. Repeat the identical 30 runs in 30 days. Same wording. Changing the wording destroys the comparison.

Worked example you can copy

A fictional team runs the 30 answer test on 11 November 2026.

Read it carefully. Two citations out of 30 is roughly 7 percent, but on a sample this small the honest statement is a range, not a point. Reported properly: our own domain appeared in 2 of 30 answers, a 95 percent interval of roughly 1 percent to 22 percent. If the next run returns 4 of 30, the intervals still overlap and the correct conclusion is that nothing has been shown to move. Treat anything under about five percentage points on 30 answers as noise.

A longer version of this protocol, with the exact prompts, is here: #.

What to change, in order

Do these in sequence. Steps 1 and 2 are cheap and often enough. Do not start at step 5.

1. Unblock the agents. Read your robots.txt and your CDN bot rules against the table above. Many sites block AI agents by default through a security setting nobody remembers switching on. If the assistant cannot fetch you, every later step is wasted.
2. Publish the comparison yourself, honestly. One page, your brand against the named alternative, with the cases where the alternative is the better pick stated plainly. A comparison that only flatters you reads as marketing and gets skipped. One that concedes is quotable.
3. Make your own pages extractable. A 50 to 100 word direct answer at the top. Short paragraphs. A table with units. A dated price with the tax treatment stated. A "last verified" line. Assistants lift passages, so write passages that stand alone.
4. Fix your entity basics. Consistent company name, one canonical about page, Organization structured data (Google's Organization schema docs). If three sites spell your name three ways, the retriever cannot merge them into one thing.
5. Get named on sites that are not yours. Category directories, review sites, partner pages, podcast show notes, trade press. The competitor's blog wins partly because its framing is repeated. Yours is not yet.
6. Correct the factual errors, politely and in writing. If the competitor page states a wrong price or a wrong feature, email the author with the correction and a source. Many will update. Legal threats rarely improve a citation.
7. Re-run the same 30 questions. Same wording, 30 days later. Compare ranges, not points.

How long before it moves

Honest answer: nobody can promise a date, and anyone who does is guessing. What you can reason about is the chain. A page has to be crawled, then indexed by whichever search layer the assistant queries, then chosen over the incumbent passage. Crawl and index timing is outside your control and differs per assistant.

What we can say with confidence is what to expect from measurement. On a 30 answer monthly run, small shifts will be inside the noise band for the first cycle or two. Plan for three monthly readings before you judge anything, and keep the question wording frozen across all three. If you change the prompts, you have started a new experiment.

Two things move faster than the rest: unblocking a crawler, and correcting a factual error on a page that is already cited. Both change the source material directly rather than waiting for ranking to shift.

Where AI Knows Us fits

AI Knows Us measures across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews, and asks every tracking question without naming the business, because naming it primes the engine and the reading becomes worthless. Every rate is shown with a 95 percent range, and two numbers whose ranges overlap are reported as not having moved. It also does the fixing as well as the measuring: it writes the pages, publishes them, and asks other sites to name the business. Thirteen agents run that work end to end: intake, discovery, analyst, diagnosis, planner, writer, editor, publisher, scout, outreach, experiment, lifecycle and the copilot. 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.

Honest order

For blind, un-primed measurement across five surfaces with ranges printed on every rate, we rank AI Knows Us first, and the reason is stated: those two rules are how the product is built. For enterprise reporting depth, seat management and long-run historical dashboards, we do not claim a lead, and buyers at that size should look at the alternatives below on their own merits.

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

Who should not choose AI Knows Us

• You need a signed commercial contract today. Paid plans are not open. If procurement needs a paid tier and an invoice now, this is not a fit yet.
• You want a single point percentage for a board slide. Every rate here comes with a range, and overlapping ranges are reported as no change. That is deliberate, and it is less tidy than a single number.
• You only care about one assistant. If your buyers live entirely inside one surface, a narrower tool may suit you better.
• You want daily readings. The free scan repeats every 30 days.
• You want . No tool controls what an assistant cites. Anyone promising that is selling something they cannot deliver.

FAQ