Perplexity Is Wrong About My Company: What to Check

Perplexity is a search-first engine. For most questions it runs a search, pulls a set of pages into the model's context, and writes an answer from them. It cites whole sources by number, not specific sentences. That design is useful to you: the evidence for most answers is sitting right under the text.

A wrong fact about your company usually enters at one of four stages:

• Access. PerplexityBot could not fetch your site, or fetched it and found nothing readable. Other pages became the only sources about you.
• Retrieval. Perplexity's search pulled in an outdated review, an old press release, a directory listing, or a page about a company with a similar name.
• Generation. The model combined two true facts into a false one, or attached a competitor's detail to your brand.
• Parametric memory. The model answered part of the question from training data, which is frozen at a cutoff date, and no retrieved page corrected it.

Each stage has a different fix. If you are unsure whether you are looking at an invented fact or an old one, our guide on telling a hallucination from an outdated answer walks through the difference.

Start with the answer itself, because it tells you where to look.

1. Ask Perplexity the question that produced the wrong fact. Ask it logged out if you can, so your history does not shape the query.
2. Note which sentence contains the error and which citation numbers sit next to it.
3. Open every cited page, not only the ones beside the wrong sentence. Search each page for the wrong fact: the price, the founding year, the product name, the headcount.
4. Record what you find for each page: the URL, whether it states the wrong fact, and the date it shows.
5. Ask the same question two or three more times, and try a reworded version. Retrieval runs per query, so a different wording can pull different sources.

Then match what you found to the likely cause.

One caveat: a citation is not proof of influence. Research on how language models cite (arXiv 2606.28358) found that citation output can be changed without changing the sources, so a page can shape an answer without being cited. Treat the citation list as the best lead you have, not the full story.

If your own site is missing from the citations, check access before anything else. Access failures are silent: nothing in your analytics tells you a crawler was turned away.

robots.txt

Perplexity documents PerplexityBot as its search indexing crawler and states it is not used to crawl content for AI foundation models. Blocking it removes your pages from Perplexity's index. Check that your robots.txt does not disallow it, directly or through a blanket rule aimed at AI bots. A minimal allow looks like this:

User-agent: PerplexityBot
Allow: /

Perplexity-User is different. It fetches pages when a user asks, and Perplexity says it generally ignores robots.txt rules. So robots.txt controls your presence in the index, not every fetch Perplexity makes.

Firewall and CDN

A firewall, CDN or bot-management tool can block a crawler that robots.txt allows. The result is a 403 or a challenge page, and the crawler gets nothing. Cloudflare changed its defaults on 15 September 2026: Training and Agent crawler categories are blocked by default on ad-displaying pages for new domains, new sites on existing accounts and existing free-tier customers, while Search stays allowed by default. If you use Cloudflare, open your bot settings and confirm how Perplexity's crawlers are treated, rather than assuming. The AI Crawler Checker requests a page as each AI crawler and shows which ones your firewall turns away.

JavaScript rendering

Independent testing in June 2026 found PerplexityBot does not execute JavaScript. If your pricing, product names or company facts only appear after scripts run, PerplexityBot sees an empty shell. Fetch the page with JavaScript disabled and check that your key facts are in the raw HTML. The Perplexity visibility audit checklist covers these checks in more detail.

When a third-party page carries the wrong fact, that page is the thing to fix. Perplexity is reading it, and it will keep reading it until it changes.

Sort the cited pages by who controls them:

• Pages you control. Your site, your profiles on directories and marketplaces, your app store listings, your social bios. Update these yourself, today.
• Pages you can request changes to. Partner pages, review sites, industry directories, old press coverage. Most have a contact or correction route. Send the specific wrong fact, the correct fact, and a link to where the correct fact is stated.
• Pages you cannot change. Forum threads, archived articles, scraped copies. Here the move is to make sure better, current sources exist for the same question.

The evidence shows a strong association between being named on the sites engines cite and appearing in their answers. It does not promise an effect. Still, when Perplexity keeps citing a handful of third-party pages for questions about you, getting accurate information onto those pages is the most direct route into the answer. Our page on correcting Perplexity by fixing the sources it cites goes further on this.

While you are reading cited pages, look for anything odd: hidden text, instructions addressed to AI systems, or claims about your brand that read like they were planted. Research on LLM search agents (arXiv 2606.16821) shows manipulated content inside the result set can change what an agent endorses, and vulnerability varies by model. If you find this on a page that mentions you, document it and report it to the site owner.

Be clear-eyed about this before you plan the work.

In your control:

• Whether PerplexityBot is allowed in robots.txt.
• Whether your firewall or CDN lets it through.
• Whether your key facts are in the HTML before JavaScript runs.
• Whether your own pages state each important fact once, correctly and consistently. Contradictions between your pricing page, about page and footer give any engine two answers to pick from.
• Whether you ask third-party sites to correct what they publish about you.

Not in your control:

• Which pages Perplexity retrieves for a given question. Retrieval is decided by the operator, per query.
• Which model writes the answer. Secondary reports as of September 2026 describe Perplexity's free tier as an automatic router and its paid plans as offering a menu of models, so two users can get different answers.
• What the underlying model learned in training.
• How the model words or combines what it reads.

The one on-page step with a positive experimental signal is stating a specific fact (a price, a number, a one-line description of what you do) on a page that can be fetched. That gives the engine a sentence it can quote. It does not guarantee the engine will quote it.

Work in this order. Each step removes a cause, and the early steps are fast.

1. Record the error. Save the question, the full answer, the citation list and the date. You will need it to confirm the fix later.
2. Fix your own pages. If your site states the wrong or outdated fact anywhere, correct it everywhere. Remove contradictions between pages.
3. Check access. Confirm robots.txt allows PerplexityBot, test that your firewall returns a normal page to it, and confirm key facts are in the raw HTML.
4. State the correct fact plainly. Put it in a sentence on a page that answers the question directly, such as a pricing page or an about page.
5. Work through the cited third-party pages. Update the ones you control, then request corrections on the rest, starting with the pages cited most often.
6. Re-check after a few weeks. Ask the same questions again and compare against your record. Pages need to be re-crawled before a change can show up, and nobody publishes how long that takes.

If ChatGPT or Google has the same wrong fact, the source is often shared. Our guide for when ChatGPT has wrong information about your business follows the same logic for OpenAI's crawlers. To see what Perplexity says before you start, run a free check of what Perplexity says about you.

These are the moves that waste time or make things worse.

• Adding schema or an llms.txt file. Neither has credible evidence behind it as a way into AI answers. Google says structured data isn't required for generative AI search, and no engine documents reading llms.txt. Existing markup does no harm, but it is not the fix.
• Rewriting pages in an "AI-friendly" style. The best-designed studies of these rewriting techniques, including C-SEO Bench, found them largely ineffective and often harmful. Write the fact clearly for people.
• Blocking Perplexity-User to stop the error. Perplexity says it generally ignores robots.txt, and blocking your own site removes your correct version from the conversation.
• Assuming robots.txt is the whole story. A firewall can block what robots.txt allows.
• Testing through an old API integration. Perplexity retired its Sonar Chat Completions endpoint on 27 September 2026, and web search is opt-in on the Agent API. A test without web search returns an ungrounded answer that tells you about the model, not about what Perplexity users see.
• Planting content aimed at the AI. Hidden text and instructions for crawlers are manipulation. They put your reputation and your search presence at risk.
• Checking once. Answers vary by query and model. Ask several versions of the question before you decide what Perplexity believes.

Can I ask Perplexity to correct a wrong fact directly? Nothing in the evidence we rely on documents a correction channel that changes Perplexity's answers about a brand. The reliable path is the one you control: fix your own pages, make sure PerplexityBot can read them, and get the third-party pages it cites corrected. For the wider question across vendors, see whether you can [ask AI companies to correct brand information](/questions/ask-ai-company-correct-brand-info).

Why does Perplexity give different answers to the same question? Retrieval runs per query, so small changes in wording can pull different sources into the answer. Secondary reports as of September 2026 also describe Perplexity routing free users automatically and letting paid users choose among several models. Different sources plus different models produce different answers. Test several phrasings, ideally logged out, before concluding what Perplexity says about you.

If I block PerplexityBot, will Perplexity stop mentioning my company? No. Blocking PerplexityBot removes your pages from Perplexity's index, but Perplexity can still cite other sites that mention you, and Perplexity-User generally ignores robots.txt. Blocking usually makes accuracy worse, because your own correct version drops out while reviews, directories and old articles remain. Allow it and make your pages readable instead.

Will updating my page fix the answer straight away? Not necessarily. Perplexity has to re-fetch the page, and then retrieve it for the relevant question. Nobody publishes how long that takes, and retrieval is not guaranteed. If the wrong fact lives on a third-party page Perplexity keeps citing, your own update may not change the answer until that page changes too. Re-check after a few weeks.

How do I stop this happening again? You cannot stop an engine from making mistakes, but you can reduce the openings. Keep each key fact stated once and consistently on your site, keep it in the raw HTML, keep crawler access open, and keep third-party listings current. Then check answers on a schedule. Our page on whether you can [stop AI making up brand information](/questions/stop-ai-making-up-brand-information) covers prevention in more depth.