A buyer asks Perplexity which vendors to look at in your category. Four names come back. None of them are yours, and every source in the citation strip belongs to someone else.
Most advice about this is ChatGPT advice with the name swapped out, and we have written the ChatGPT version separately. That misses the thing that matters, because the two work differently in one specific way, and that difference changes both why you’re missing and how quickly you can do anything about it.
Why Perplexity works differently from ChatGPT
Perplexity looks things up. ChatGPT can answer from what it already learned. That single difference drives everything else: Perplexity has almost no trained memory of your category to fall back on, so when the live web holds nothing solid about you, there’s nothing for it to find and nothing for it to say.
Every AI answer is built from two things, which we have set out in full in where ChatGPT gets its information. There’s the training data baked into the model when it was built, and there’s retrieval, meaning whatever the system pulls in live at the moment somebody asks. ChatGPT leans on both. Ask it a broad question and it’ll often answer straight from memory without citing anyone.
Perplexity is built the other way round. It runs a search, reads what it gets back, and writes an answer out of those sources. The citation strip isn’t decoration; it’s the answer’s actual ingredient list.
| Perplexity | ChatGPT | |
|---|---|---|
| Primary source of an answer | Live retrieval, nearly always | Trained memory, with retrieval on specific or current questions |
| If the live web is thin on you | You’re absent | You might still be named from memory |
| Where the fix lands | The sources it can reach | The sources it can reach, plus the slow work of being known |
| How fast a fix shows | Days to weeks | Weeks to a retrain cycle |
That’s the whole argument. Retrieval is the half you can move now, and on Perplexity it’s very nearly the only half there is.
What Perplexity reads when someone asks about your category
It reads what it can fetch in the moment: pages it can crawl and render, review platforms, comparison articles, community threads. Your own site is one input among many, and rarely the loudest one. Whoever wrote the roundup that ranks today shapes the answer more than your product page does.
Sit with that for a second, because it reorders the work.
If you’re arguing internally about homepage copy while your category’s “best tools for X” roundup doesn’t mention you, you’re polishing the input that carries least weight. The pages Perplexity reaches for on a vendor question are the ones written about vendors, not by them. Review platforms, editorial comparisons, the odd Reddit thread where somebody asks the same question a buyer just asked.
We’ve written about why that happens in how reviews and third-party mentions shape AI recommendations, and the short version holds here: other people’s pages about you outrank your pages about you. It’s just how the game works, and it has been for years. AI search made it visible rather than making it true.
What this does change is the stakes. A roundup that skips you used to cost you one referral path. Now it feeds a system that a buyer treats as a summary of the whole market, which turns the same omission into an influence channel running against you.
The reasons you’re missing, in the order we find them
In our work the causes come up in a consistent order. The crawler can’t reach you. Your category label is ambiguous. Third-party evidence about you is thin. No comparison content includes you. What does exist about you is out of date. Technical causes are the rarest and the easiest; evidence causes are the commonest and the slowest.
The crawler can’t reach you
Rare, and worth ruling out first because it takes ten minutes. Check whether PerplexityBot is blocked in robots.txt, whether your key pages render without JavaScript, and whether bot protection is turning away legitimate crawlers along with the bad ones. A staging noindex that survived a launch will do it too.
Ten-minute check: fetch your own robots.txt and search it for PerplexityBot, then load a key page with JavaScript disabled and see what survives.
When this is the cause, fixing it is the fastest win available anywhere in AI search. We didn’t expect to find it as often as we do on well-run sites, and it’s almost always something nobody chose deliberately.
Your category label is ambiguous
Perplexity has to decide what you are before it can decide whether you belong in an answer. If your own site describes you three different ways across the homepage, the product pages and the about page, you’ve made that harder than it needs to be.
Pick the category name your buyers use, not the one your positioning deck invented, and say it consistently. The entity is the asset here: the system has to be able to file you somewhere before it can recommend you from there.
We tried treating this as a copywriting job on one engagement and it didn’t work, because the inconsistency wasn’t on the website at all. It was on the review profiles, the funding announcements and an old partner directory, all describing the company as something slightly different.
Third-party evidence about you is thin
The most common cause, and the one that stings. Perplexity needs sources, and if there are barely any pages about you on sites it trusts, it has nothing to assemble an answer from. Review profiles with three reviews, no analyst coverage, no press, no community presence: that’s a vendor the system can’t describe with any confidence, so it names one it can.
No comparison content includes you
Vendor questions pull comparison formats disproportionately, which is the same reason getting recommended by ChatGPT turns on the same pages. “Best X”, “X alternatives”, “X vs Y”. If your name appears in none of those pages anywhere on the web, you’re absent from the format the answer is most likely to be built from.
What exists about you is out of date
The gentlest failure and the most annoying. You’re named, but with pricing you retired, a feature set from two versions ago, or a founder who left. Perplexity found something real and old, and it has no way to know which version of you is current unless the current one is easier to find.
The check that tells you which one it is
Run the same three questions a buyer would ask, in a clean session, and read what comes back rather than whether you appear. What Perplexity cites tells you where the answer is being formed. If your competitors’ names arrive via review sites and roundups, that’s where your gap sits, not on your own site.
Ask a broad category question with no vendor named. Ask a narrower one with a use case attached. Ask directly about your company.
Then read the citation strip on each, and write down the domain of every source. That list is your diagnosis. If the sources are review platforms and comparison articles, and you’re absent from those sources, it’s an evidence problem. If your own site is cited but the details are wrong, it’s a source problem on a page you control. And if nothing about you appears anywhere, check the crawler before assuming the worst.
This is the same logic we use to find the gaps in AI search visibility more broadly, run at a single platform.
What fixes it, and how fast
Retrieval problems move quickly. Unblock the crawler or correct a page the system already reads and the answer can shift inside a fortnight. Evidence problems move slowly, because you’re waiting on other people to publish, review and mention you. Authority is the slow one, and it also decides most of the rest.
| Cause | Typical fix | How fast |
|---|---|---|
| Crawler blocked | robots.txt, server-side rendering, bot rules | Days |
| Out-of-date information | Correct the cited page or listing | Days to weeks |
| Ambiguous category | Consistent language across your own pages | Weeks |
| Missing from comparisons | Earn inclusion in roundups and alternatives pages | Months |
| Thin third-party evidence | Reviews, digital PR, coverage, community presence | Months |
The honest reading of that table is that the quick fixes are the rare ones. Most companies who ask us this question have a working site and an evidence problem, which means the answer is months of earning presence rather than an afternoon of tags.
Build the entity first. A vendor the web talks about gets described by systems that have never been asked to try very hard; a vendor nobody references stays a blank the model routes around.
A 30-minute call where we map what AI says about you and build a tailored roadmap.
If Perplexity has you, but has you wrong
Being named and being described correctly are separate problems with separate fixes. If Perplexity is citing a stale review or a two-year-old comparison, the answer changes when that source changes, and not when your site does.
Follow the citation, find the page, and get that page corrected. Where the wrong detail sits in the model’s memory rather than in a source it fetched, you’re into much slower work, which we’ve set out in why AI describes your product wrong.
