Citation Problem Or Entity Problem? Diagnosing Why AI Ignores Your Brand

A brand that ChatGPT, Perplexity and Google AI Overviews ignore usually has one of two different problems, and more content fixes neither by default. A citation problem means engines cannot retrieve or extract a usable passage from your pages. An entity problem means engines cannot tell what your brand is. Test for both before publishing anything.

This applies when your brand is absent from, or misdescribed in, AI answers to the category questions your buyers actually ask, and your pages are otherwise indexed and live.

The two-cause split is a diagnostic method drawn from TRM's audit work, not a published research finding, and real cases are often a mix of both.

What Counts As An AI Citation, And Why Ignored Brands Have Two Different Problems

An AI citation is a linked source an answer engine shows or uses to support a statement in its answer. ChatGPT, Perplexity and Google AI Overviews all produce answers that lean on outside pages, and the pages they lean on are the citations.

A citation is not a ranking. A page can sit in the top three on Google and never appear as a source in an AI answer. It is also not a mention. A mention is your brand named in the answer text without a link to your site, and the two can happen independently. An engine can name you and never link you, or link you and never name you. Both count as visibility of a kind, but only one of them sends a reader to your pages.

Ranking helps, but it does not settle the question. In one publisher's 10,000-citation study, a number one Google ranking was cited by an AI engine about 60% of the time, against about 11% for page two or lower. That publisher sells AEO software and has not published its per-engine samples or method, so read the figures as one study's result and not as a law. The pattern is still useful. Ranking raises your odds, and plenty of number one pages go uncited anyway.

Why does any of this matter? Because a buyer who asks an engine for the best option in a category gets a short list, not ten blue links. Brands on that list are in the consideration set. Brands off it are not, and the buyer never sees the gap.

When a brand is missing from that list, teams often reach for the same answer, which is more content. Sometimes that is right. Often it is the slowest possible fix, because the real cause sits somewhere content volume does not touch. There are two causes worth separating.

• Citation problem: Engines cannot retrieve or extract a usable passage from your pages, so they cite someone else's.
• Entity problem: Engines cannot tell what your brand is, who it serves or how it differs from similar names, so they leave it out of the answer.

The rest of this article is a method for telling them apart. You run a test query set, read the sources behind the answers, and follow a decision sequence that points at one cause or the other. Each cause then gets its own fix path.

How AI Engines Pick The Sources They Cite

In our reading, engines appear to work in two steps. First they retrieve candidate pages for the question. Then they pick passages from those pages to support the answer and attribute them. A brand can fail at either step, and the two failures look different in a test log. That is the whole reason the split is useful.

Start with the one rule that is written down. Google's own AI features documentation states that a page is eligible to appear as a supporting link in AI Overviews and AI Mode if it is indexed and eligible to be shown in Google Search with a snippet, with no additional requirements. That is a low bar, and it matters because it rules out a whole family of fixes. The same page also states that no special schema.org markup or AI-specific file is needed. If someone sells you a markup package as the way into AI Overviews, ask them to show you where Google says so.

Eligibility is the floor, not the ceiling. Being allowed to appear does not mean being chosen. Among eligible pages, our working assumption is that engines favour passages that answer the question most directly, and that is where a citation problem would live. Separately, the engine has to know what your brand is before it can name it or attribute a passage to it, and that is where an entity problem lives. If you want the full tactics list after the mechanism, our guide on how to get cited in AI answers covers it.

The engines also differ in how much of this you can see. That matters for testing, because you can only diagnose what you can observe.

The speed column deserves a hedge. The timelines come from vendors, not from independent research, and they describe typical behaviour rather than a guarantee. Treat the order as the useful part. Perplexity and AI Overviews tend to reflect new pages sooner, and ChatGPT tends to lag.

Step 1: Build A Test Query Set That Exposes The Gap

Run five to ten buyer-style questions across three engines and log engine, appearance, competitor named and sources shown. That is the whole method, and it takes an afternoon. Here is how to do it properly.

1. Write the questions a buyer would ask. Use category questions such as the best tools for a job, comparison questions, and problem questions. Skip your brand name. A buyer who already knows you is not the buyer you are losing.
2. Keep the wording stable. Write each question once and reuse it verbatim in every engine and every later round. If the wording changes, you cannot tell whether the result changed or the question did.
3. Run every question in ChatGPT, Perplexity and Google AI Overviews. Use a fresh session each time so earlier answers do not colour later ones. Three engines give you enough spread to see whether a gap is general or specific to one.
4. Log each run in the same shape. Record the prompt, the engine, whether your brand appeared, which competitors were named and which sources were shown. A spreadsheet is enough.
5. Save the evidence. Copy the answer and the source list as they appeared. Answers change, and you will want the original when you compare rounds.

The log is only useful if every row has the same columns. Here is a template.

Why five to ten? Fewer than five and one odd answer skews the picture. More than ten and you spend the week logging instead of reading. The point of this round is to find a pattern, and a pattern shows up quickly when the gap is real.

Why avoid the brand name? Because a brand-name query tests whether the engine knows you exist, and most engines will answer that. Buyer questions test whether the engine reaches for you when it does not already have your name in front of it. That is the question that matters.

One more design point. A single question can be worded five ways and return five different answers. If your set mixes a few rephrasings of the same intent, you learn whether a gap is stable or wording-sensitive. Our guide to prompt variation testing tools covers that design problem in more depth.

Step 2: Read The Cited Sources, Not Just The Answer

Most people read the answer and stop. The answer tells you whether you were named. The sources tell you why.

For every query where your brand is absent, pull the cited sources and sort each one into a type. Perplexity makes this easy because its numbered citations show which domains an answer used. For the other engines, use whatever source list or supporting links the answer shows.

Why does this step matter so much? Because brands tend to assume engines cite brand websites. The evidence points the other way. In one 10,000-citation study across 57 queries and four AI engines, owned vendor sites accounted for under 4% of total citations. The same study found third-party sources lead cross-platform citations, with Reddit, Wikipedia, YouTube and Forbes at the top. Again, the publisher sells AEO software and does not disclose per-engine sample sizes or its counting method. This is one publisher's study, so use it as a prompt to look at your own logs and not as a benchmark to hit.

What does that mean for reading your source list? If your own pages almost never appear and outside pages dominate, then rewriting your homepage will not move the result. The fix lives on pages you do not control. If the sources are mostly pages like yours and yours is missing, the fix lives on your site.

Reading which URL got cited, and not only which domain, makes this step sharper. Page-level tools help here, and we compare them in our piece on page-level citation attribution tools.

Finish the step by writing one line per absent query. Name the dominant source type and name two or three of the domains. You now have the evidence the next section needs.

The Diagnostic Decision Sequence: Citation Problem Or Entity Problem

This is the core of the method. Follow it with your query log open.

A label before you start. No source defines an entity problem as a separate diagnosis, so the mapping below is TRM's inference from audit work. It is a reasoned framework you can test against your own logs, not a measured result. Treat each row as a hypothesis with a first check, and then check it.

Work through four questions in order.

1. Is the page indexed and snippet-eligible? If not, nothing else matters yet. Fix that first, because Google's own rule makes eligibility the entry ticket.
2. Are your pages retrieved or cited for other queries but not this one? Then the engine knows your pages and finds this one unusable for this question. That points at passage-level quality, which is a citation problem.
3. Does the engine name your brand wrongly, name it as something else, merge it with another company, or never name it even where third parties discuss it? That points at an entity problem.
4. Are competitors cited from third-party pages that never mention you? Then the problem is presence in the sources engines rely on, which is part of the entity picture.

Now, three worked examples. All are hypothetical logs, written to show the reasoning. None is a client result.

• Hypothetical log A: A project management tool appears in no answer, but Perplexity cites its pricing page for a pricing question. The engine knows the brand and the page. The category question lacks a usable passage on any page. This reads as a citation problem, and the first check is where the category answer sits on the relevant page.
• Hypothetical log B: A security startup is named in ChatGPT as a consultancy, and Perplexity merges it with a company that shares half its name. Its pages are retrieved but the descriptions do not match. This reads as an entity problem, and the first check is whether the name, category and description are identical across the site and every profile.
• Hypothetical log C: An analytics vendor is absent everywhere, and the cited sources are three comparison articles and a community thread, none mentioning the vendor. The vendor's own pages are fine. This reads as a presence problem in third-party sources, and the first check is which of those publications cover the category.

Notice that none of the three would be fixed by publishing another blog post. That is the point of running the test before the plan.

If It Is A Citation Problem: Signals And The Fix Path

A citation problem shows up as indexed pages that never appear among the sources for the query, even though the brand is recognised. The first fix is to confirm the pages are indexed and can show a snippet.

Look for three signals in your log.

• Pages are indexed but never cited: The engine can reach the page. It simply does not pick it for this question.
• The answer is buried: The passage that answers the question sits low on the page, or inside a long block of text with no clear start.
• Weaker brands win: Competitors with less recognition get cited because their passage answers the question directly and yours does not.

Fix it in this order.

1. Confirm indexing and snippet eligibility. Google's documentation says a page needs to be indexed and eligible for a snippet in Search to appear as a supporting link. Check crawl access and indexing on the pages in question first. Our page on technical SEO and AI readiness covers the checks.
2. Put a direct answer near the top of the two or three most relevant pages. Do this before you touch anything else. Rewriting the pages that matter most, so the answer sits near the top, is the usual first fix for absence. Do not rewrite the whole site.
3. Make the passage self-contained. The answer should make sense if an engine lifts it alone. Name the subject in the sentence, attach the qualifier to the claim, and put units on every number. Our approach to answer engine optimization is built on this idea.
4. Re-test the same queries. Change one thing at a time and run the same set, so you know which edit moved the result.

Schema is not a requirement here. Google states that no special markup is needed to appear in AI Overviews or AI Mode. Structured data can still serve other purposes, but do not expect it to rescue a page whose answer is buried.

For the technical blockers specific to Google, our guide on increasing visibility in Google AI Overviews goes through extraction problems and how to spot them.

One last point on this fix path. A citation problem is the easier of the two to repair, because every piece of it sits on pages you own. If your log points here, you can start today.

If It Is An Entity Problem: Signals And The Fix Path

An entity problem shows up when engines describe your brand inconsistently, or leave it out of category answers even when your pages are retrieved. The first fix is one canonical description of what the brand is and who it serves.

Watch for these signals.

• Wrong or generic description: The engine describes the brand as something it is not, or in words so vague they could fit any company.
• Mistaken identity: The engine confuses the brand with a similarly named company.
• Missing from category lists: The brand is omitted from category answers even where its own pages are retrieved.
• Inconsistent facts: The brand's name, category and description differ across its site and the profiles, directories and publications engines read.

The last signal is the quiet one. Teams rarely notice that the homepage says one thing, the LinkedIn page another and the old directory listing a third. An engine reading all three has no stable picture to work from. Our piece on PR and brand messaging for AI visibility explains why inconsistent messaging is a named cause.

Fix it in this order.

1. Write one canonical description. One sentence on what the brand is, one on who it serves. Use the same category words a buyer would use. This is the text every other place should echo.
2. Repeat it everywhere. Put it on the site, on every profile, in every directory entry and in every author bio. Differences in wording, category or spelling make the picture muddier. The way language models read and resolve a brand is covered in our LLM SEO approach.
3. Earn mentions on the third-party sources your cited-source table showed. If the table pointed at industry publications and community sites, those are the places to be present. Our work on earned mentions and digital PR is built for exactly this. Not every placement carries equal weight with engines, so it helps to know which link placements AI engines already cite before you choose.
4. Check entity coverage. Some tools surface where a brand and its topics are missing, and we compare them in our review of entity and topic coverage tools.

Do not create your own Wikipedia entry. A Wikipedia presence depends on notability established by press and analyst coverage, so a brand that writes its own page is working against that rule. Earn the coverage first, and let the entry follow if it is warranted.

This fix path takes longer than the citation one, because it depends on pages and people outside your control. Teams that skip it tend to keep publishing without changing how engines see the brand.

If your log shows that your brand appears in Perplexity but not in ChatGPT, look at the entity signals first. The engines read and show sources differently, and one can resolve a brand where another cannot. A gap in only one engine is a useful clue, not a verdict.

When The Test Shows Both Problems: What To Fix First

Real logs are often messy. Some queries show buried answers, others show a muddled brand, and many show both. When that happens, you need a rule for what comes first.

If engines cannot name your brand, fix the entity first. If engines name you but never use your pages, fix the passages first.

The reasoning is simple. A cleanly extractable page for an unrecognised brand still goes uncited, because the engine has no confident idea what to attribute it to. A recognised brand with a buried answer is the opposite case. The engine already knows who you are, so a better passage can start working as soon as it is picked up.

This is a judgement call, not a measured result. Nobody has published a trial that ranks the two orders. It is the order we would try first, and your own re-tests will tell you whether it held for your brand.

Fixes can also undo themselves. A citation you earned this month can vanish next month, and you will not notice unless you look. Alerts that flag the loss are worth setting up once you have something to protect, and our review of citation loss alerts covers the options.

How Long Fixes Take And How To Re-Test

Re-run the same query set every week and judge the pattern across batches, because a single result is a snapshot. One answer from one engine on one day tells you very little. Four weekly batches that move the same way tell you a lot.

Expect different speeds from different engines. Reported timelines suggest Perplexity and Google AI Overviews can pick up new pages within days and cite them within weeks, while changes in ChatGPT and Claude can take three to twelve months. Those figures are vendor-reported and loose, so read them as a rough order, not a schedule. The practical lesson is to judge each engine against its own clock. A fix that has not shown up in ChatGPT after three weeks has not failed yet.

Keep the discipline of the first round. Same questions, same wording, same log columns. Change one variable between rounds where you can. If you rewrite a page and fix a profile in the same week, you will not know which one moved the result.

When the same query set needs tracking week after week, a tracker such as Semrush keeps the history in one place instead of a spreadsheet you rebuild each Monday.

If you would rather compare options before you commit, our roundup of tools for monitoring AI citations goes through what each one tracks.

What about traffic? A citation is a link, so it can send visits, and a citation or mention also puts your name in front of a buyer at the moment they ask. Do not expect a clean line from one to the other. Watch referral visits and branded search over the same weeks as your re-tests, and read them together. The goal of the first few rounds is movement in the log, and traffic follows once the log is moving.

Common Misreadings Of A Test Result

These mistakes send teams to the wrong fix. Each one has a cause and a correction.

• Testing with the brand name: A brand-name query only shows that the engine knows you exist. Correct it by testing buyer questions that never mention you.
• Reading one run: One answer is a snapshot and can change the next hour. Correct it by repeating the set weekly and reading the pattern.
• Assuming schema will fix absence: Google states in its AI features documentation that no special markup is needed, so absence has another cause. Correct it by running the decision sequence first.
• Treating a mention as a citation: A brand named in the answer text without a link to its site is a mention, and it does not send a reader to your pages. Correct it by logging the two separately.
• Treating citation stability as guaranteed: A source cited today can drop out tomorrow. Correct it by keeping the weekly re-test running after the first win.
• Treating the whole log as one problem: Different queries can point at different causes. Correct it by diagnosing each absent query on its own line before you decide on a plan.

If you recognise two or three of these in your own process, your last test was probably misleading. Rerun it clean before you spend money on a fix.

Where A Self-Run Diagnosis Stops And An Audit Starts

A manual test covers one query set across a few engines, and it can tell you which cause is likely but cannot confirm it across the wider set of questions your buyers ask. A prioritised fix list across many queries is where an audit earns its place, and if you are not ready for that, our free AI Visibility Report is the lighter step.