Perplexity Searches Literally. ChatGPT Rewrites Everything.

Most AI visibility advice treats Perplexity and ChatGPT as variations of the same problem. Add schema. Clean up your directories. Fix NAP consistency. These actions appear in every guide, for every platform, without qualification.

In our June 26, 2026 investigation into platform retrieval architecture (`knowledge/platform-divergence-data.md`, session 59), we found data that breaks this assumption at the mechanism level. Perplexity and ChatGPT don't just prefer different sources -- they search for content using structurally different query logic. Vocabulary precision is the right fix for one platform. It's irrelevant for the other.

The 88% / 91% split

When a buyer types a query into Perplexity, what does Perplexity actually search for?

The session 59 data, from Nick Lafferty's analysis of AI engine citation asymmetry (nicklafferty.com/blog/ai-engine-citation-asymmetry/, 2026), provides a specific answer. Perplexity retains approximately 88% of the original prompt words when it runs its retrieval query. If a buyer types "burst pipe emergency plumber near me," Perplexity searches for sources containing language very close to "burst pipe emergency plumber near me."

ChatGPT runs the opposite logic. Processing the same query, it generates approximately 91% unique sub-queries -- sub-queries that don't appear verbatim in the original prompt. "Burst pipe emergency plumber near me" decomposes into something like: "cost of emergency plumber," "licensed plumber after hours," "how to shut off main water valve," "plumbing emergency reviews," "24-hour plumber availability." Five or ten queries, most of them structurally unrelated to the original phrasing.

This is not a difference in degree. It's a structural difference in how each platform searches. Perplexity behaves like a searcher who copies the query into a search bar. ChatGPT behaves like an analyst who breaks the question into every angle it contains before going to look for answers.

A corroborating data point from the same session 59 research: Moz's study of 68,313 keywords and 1.3 million citations (February 2026) found that only 12% of Google AI Mode citations matched URLs in the top-10 organic results for the same keyword. AI Mode uses the same fan-out architecture as ChatGPT -- it generates related sub-queries and aggregates citations across all of them. Traditional organic rankings predict AI citations for these platforms at roughly the same rate as a coin flip.

What this means if your Perplexity score is low

The 88% literal retention rate has a direct content implication: if your page doesn't use the words your buyers actually type, Perplexity won't retrieve it for their queries.

This sounds obvious, but most service businesses use professional language on their pages and buyer language in their heads. A plumbing company writes "residential emergency plumbing services available 24/7." A homeowner types "burst pipe plumber near me" or "water everywhere call plumber." These phrases don't match. Perplexity's retrieval skews toward the second set; pages written with the first set don't clear the vocabulary threshold.

Our June 22, 2026 investigation into incident-framed query coverage (`methodology-recs/2026-06-22-incident-framed-b-queries.md`, session 55) documented the scale of this gap. The 5W HVAC & Plumbing AI Visibility Index (350,000 queries) found that 75% of homeowners typed "keyword shorthand identical to Google search" -- not conversational prompts. "AC stopped working [city]." "Burst pipe [city]." "Emergency electrician tonight." Not "I'm experiencing an issue with my HVAC unit and looking for qualified service providers."

If your service page doesn't contain the shorthand -- the exact incident description a buyer types under urgency -- Perplexity's literal retrieval passes over it.

The vocabulary gap is usually larger than businesses expect. A useful diagnostic: read your service pages out loud and ask whether a homeowner, calling at 11pm with a broken furnace, would have typed those words. If the answer is no, that gap is likely driving your Perplexity absence.

The practical fix is also more tractable than most content improvements. You are not changing what you say -- you are changing the vocabulary you say it in. One paragraph per service page that uses incident-framing ("When your AC stops working in [city]...") alongside professional descriptions costs thirty minutes and creates a direct vocabulary match for Perplexity's literal retrieval.

What this means if your ChatGPT score is low

The 91% unique sub-query generation means vocabulary precision is not the fix for ChatGPT. A page containing every possible buyer phrase still reaches only one angle of the decomposed query space.

ChatGPT's retrieval is a breadth problem, not a vocabulary problem. Each sub-query it generates searches a separate angle: how much it costs, what the emergency indicators are, which providers have reviews, how to compare options, what to check first. A business that creates content -- or achieves directory presence -- across multiple of these angles creates more opportunities to appear across ChatGPT's sub-query passes.

This is consistent with GPT-5.5's behavior shift, which we documented in our June 16, 2026 fix plan stratification rec (`methodology-recs/2026-06-16-gpt55-chatgpt-citation-shift.md`, session 49). Under GPT-5.5 Instant -- ChatGPT's default model for free users -- the engine now uses the `site:` operator to target brand-specific domains in only 12.6% of its web search queries, down from 40.5% under GPT-5.4. ChatGPT stopped proactively seeking brand vocabulary. It now searches topically, and the retrieval goes to whatever is topically present in the result pool.

Third-party directory listings work better here than vocabulary optimization, because each platform covers a different topic angle independently. A business listed on Yelp (reviews), BBB (trust and longevity), and Angi (contractor comparison) has three separate entry points across ChatGPT's sub-query space -- the reviews sub-query finds the Yelp entry, the comparison sub-query finds Angi, the trust sub-query finds BBB. Each listing hits a different angle. A single optimized page hits one.

The 46x citation rate gap shows why the mechanism matters

The platform divergence data from Averi's 2026 B2B SaaS Citation Benchmarks report (680 million citations, independently confirmed by two separate studies using different methodologies) shows Perplexity cites brands in 13.05% of relevant responses. ChatGPT cites brands in 0.59% of responses. A 46x difference in citation rates (`knowledge/platform-divergence-data.md`, session 42 major update, 2026-06-09).

This isn't a measurement quirk. It reflects the structural architecture above. Perplexity runs live retrieval on every query and includes cited sources by design -- every response contains the sources it retrieved. When vocabulary alignment is correct and the business is in the source pool, the citation mechanism is direct. ChatGPT defaults to parametric knowledge (training data) and only appends web search citations when it decides a live lookup is necessary. Under GPT-5.5, that decision happens less often, not more.

The practical implication: improving vocabulary alignment on Perplexity has a traceable path to measurable citation improvement. The 13% citation rate makes that progress visible. ChatGPT citation rates are structurally lower -- the fix is worth pursuing, but it operates at scale through distribution rather than through precision.

Which gap to diagnose first

The 88%/91% split suggests a practical diagnostic sequence.

If your Perplexity score is low, check vocabulary first. Pull the exact query terms from your Signal Check audit and compare them against the language on your service pages. If there's a vocabulary mismatch between how the queries are phrased and how your pages are written, that's your primary Perplexity gap. The fix is page-level and fast.

If your ChatGPT score is low, check distribution. How many distinct topic angles does your content -- including third-party listings -- collectively cover for your category? A business with one well-optimized page and two directory listings has limited surface area across ChatGPT's sub-query fan. A business with category presence on six third-party platforms, each covering a different aspect of the service, has much more.

A Sourcepull audit identifies which platform is driving your citation gap and which query types are failing. That platform-specific diagnosis determines whether vocabulary precision or distribution breadth is the right investment -- and in which order.