Why One AI Engine Cites You While Another Ignores the Same Page

You publish one strong article. Two weeks later a colleague asks ChatGPT a question in your category and your brand shows up in the answer. You feel good, so you ask the same question in Perplexity, then Gemini, then Claude. Nothing. Same page, same topic, three silences.

This is the part of AEO most guides skip. They treat "AI search" as one channel with one set of rules. It is not. Each engine runs its own retrieval pipeline, reads its own index, and applies its own scoring logic before a single source makes it into an answer. A page tuned for one can be invisible in another for reasons that have nothing to do with quality. If you want to know the mechanics of getting quoted in the first place, start with our guide on how to get your brand cited by ChatGPT, Claude, and Perplexity. This piece goes one level deeper: why the four major engines disagree about the same URL, and what to do about it.

The retrieval layer nobody optimizes for

Before an engine can cite you, it has to find you, and "find you" does not mean the same thing across engines. Citation is a two-part process: retrieval (pull candidate pages from an index) and selection (decide which few actually appear). You can write the best answer on the internet and still lose at step one if you are not in the index that engine reads.

Here is where each engine looks:

That single table explains most cross-engine mysteries. If you rank on Google but have thin Bing presence, Gemini may quote you while ChatGPT never sees you. If your page is a broad overview with no crisp, checkable claims, Perplexity has nothing clean to attach a citation to. Getting your content into the feeds these crawlers actually ingest is a discipline of its own, which is the entire point of an AI-native content feed engine.

How ChatGPT picks citations

ChatGPT with Search runs a three-stage pipeline: retrieve from the Bing index, rank candidates, then select a final few. The attrition is brutal. Analyses of ChatGPT citations report that roughly 85% of retrieved pages never appear in the final answer, which means being retrieved is table stakes, not a win.

The ranking signals that survive that cut are consistent across studies:

• Domain authority carries real weight. Pages that rank #1 in Google have been found to earn citations around 3.5x more often, and the highest-authority tier of sites (scoring 97 to 100) averaged roughly 8.4 citations against 1.6 for sites below 43 in one analysis by Profound's citation research.
• Freshness matters more than people expect. Content updated within the last three months has been reported as about 3x more likely to be cited. A "published 2022, never touched" page decays out of the running.
• Structure decides extractability. A large share of citations come from the first third of a page. Clear H2 and H3 headings, a direct answer in the first sentence of each section, and FAQ blocks all raise the odds. This is the same structural logic behind the page structure that actually gets cited by AI assistants.
• Earned media beats owned media. ChatGPT leans on third-party coverage. A frequently cited figure is that around 85% of top-of-funnel brand mentions in AI search come from third-party content, not your own domain.

One more quirk worth planning around: ChatGPT's browsing is inconsistent. Studies have found under 10% citation consistency across five identical prompts, so a single test proving you are "not cited" can be noise. Test in volume, not once.

How Perplexity picks citations

Perplexity is the most citation-forward engine by design. It runs its own crawler and is built to retrieve and attach sources for almost every claim, because traceability is its core product promise. That changes what wins.

Perplexity does not reward broad, hedged prose. It rewards a specific sentence it can point at. If your page says "many companies see improvement," there is nothing to cite. If it says "teams that added FAQ schema saw citation rates rise in our 40-account sample," that is a citable, traceable claim. Write in checkable statements: numbers, named methods, concrete comparisons. Because Perplexity attaches sources aggressively, the ceiling for being cited is higher here than anywhere else, but only if your content gives it clean, attributable units.

Perplexity also weights recency and query-answer match tightly. A page whose H2s mirror the exact question phrasing, followed immediately by a one-to-two sentence answer, gives Perplexity the cleanest possible extraction target.

How Gemini picks citations

Gemini grounds its answers in Google's index and the properties Google already trusts. If ChatGPT is Bing-shaped, Gemini is Google-shaped, and everything you know about entity authority on Google applies.

Two signals dominate. First, structured data. Gemini heavily weights schema markup and Knowledge Graph entities, and pages with comprehensive, valid schema get preferential treatment as candidates it can confidently ground against. Second, existing Google authority. Content Google already ranks and associates with a clear entity is content Gemini reaches for. Gemini also does not cite every sentence; it surfaces sources mainly when it lifts or closely matches a passage, so it favors pages that own a specific answer cleanly rather than pages that touch a topic loosely.

Practically, Gemini is where your schema hygiene and entity clarity pay off. A clean Organization entity, FAQPage markup on your answer content, and consistent naming across the web all raise your grounding odds. If you have not yet published a machine-readable map of your site, our free llms.txt generator is a fast first step toward making your content legible to crawlers.

How Claude picks citations

Claude behaves differently enough that a page tuned only for Google-shaped engines can miss it entirely. Claude decides on its own when a prompt needs a web search, runs that search against a live index that is not Google's, and then cites the sources it used. Anthropic documents the mechanism in its web search tool documentation. Claude is also unusually cautious about unverified claims and unusually fond of explicit, well-structured sources.

Independent citation studies of Claude in 2026 found a distinctive pattern that is worth internalizing:

• Roughly 56% of cited URLs sat under a /blog/ path, and about 47% used listicle-style paths such as /best-, /top-10-, /alternatives, or /vs-.
• Around 24% of cited URLs contained a year token like 2025 or 2026 in the URL itself.
• Claude cited almost no mainstream news outlets or major social platforms, favoring practitioner and SaaS company blogs instead.
• Citations spread across a very long tail: in one study the top 10 domains accounted for just 9.5% of all citations, and about 32.9% of cited domains appeared only once.

The takeaway is that Claude rewards the exact opposite of a thin, anonymous corporate page. It surfaces specific, dated, practitioner-authored content with a clear point of view. That long-tail behavior is also good news for smaller brands: you do not need to be a top-10 domain to get cited by Claude, you need to own a specific answer.

One page, four verdicts: the cross-engine playbook

You do not need four versions of a page. You need one page engineered to clear every engine's bar at once. Here is how the requirements stack.

Read down the columns and a shared spec emerges:

1. Fix retrieval on both indexes. Confirm you are indexed by Bing and Google, and that your key pages are not blocking AI crawlers in robots.txt. This is the cheapest, highest-impact fix, and it is the one most teams never check.
2. Lead every section with the answer. A 40 to 60 word direct answer under a question-shaped H2 satisfies ChatGPT's structure preference, Perplexity's extraction target, and Claude's appetite for explicit content in one move.
3. Write in checkable claims. Replace vague assertions with numbers, named methods, and comparisons. This is what earns Perplexity and Claude citations and what makes your content quotable rather than summarizable.
4. Ship valid schema and a clean entity. FAQPage and Organization markup plus consistent naming is your Gemini insurance policy.
5. Keep it dated and fresh. Update timestamps, cite current data, and where honest, include the year. Freshness moves ChatGPT and Claude, and dated URLs correlate with Claude citations.
6. Earn third-party mentions. Owned content alone underperforms in ChatGPT. Roundups, guest analysis, and review-site presence feed the earned-media signal every engine partly relies on.

Do all six on one URL and you stop guessing which engine will quote you. You built the page each of them is looking for. For a live example of this compounding across a real content program, see the FastTrackr AI case study, and for the mechanics of how the whole system runs end to end, how OnlyAEO works walks through it.

How to measure this per engine

If the engines cite differently, you have to measure them separately. A single "AI visibility" number hides the exact gaps you need to close. Track citation share engine by engine so you can see, for example, that you are at 20% in Perplexity and 0% in Gemini, which tells you your schema and Google entity work is the bottleneck, not your writing.

Build a fixed prompt set of the real questions buyers ask in your category, run each prompt across all four engines on a schedule, and record whether you were cited and by which URL. Because ChatGPT browsing is inconsistent, run each prompt several times and use the rate, not a single pass. Our full method for turning this into a board-ready metric is in how to measure your brand's AI citation share across LLMs. For a broader framing of the discipline, Semrush's answer engine optimization overview and Frase's complete AEO guide are useful references to hand a skeptical colleague.

Once you can see per-engine share move, the work stops being mysterious. You publish, you measure, you find the one engine you are losing, and you fix the specific signal it cares about.

FAQ