Every “how to get cited” guide is written for one surface at a time, because that is how the tracking tools are built: a ChatGPT tracker, a Perplexity checklist, a Gemini audit. The question a B2B SaaS buyer actually asks is different: which of these surfaces matters for my category, where do the answers come from, and where am I missing? This is the only place on the internet that attempts the cross-surface answer for B2B — methodology first, a clearly labeled sample run, and the known retrieval mechanics per surface, each checked against primary sources.

Before the data, the honest framing of what this page is and is not. The sample section below shows illustrative, clearly labeled sample data — one category, a demonstration of the run shape, not a measurement. The measurement harness that will produce live, scheduled, multi-month data is under construction. This page is the launch version, and it is a living page: every time the harness completes real runs, the sample table is replaced or supplemented with live data and the update log below gets a dated entry. “Updated ” here is a promise, not a badge — the log is where you can check whether it is kept.

Why surfaces differ: retrieval mechanics, verified per surface

The surfaces do not cite differently because their models have different tastes. They cite differently because their retrieval pipelines are different systems, pointing at different indexes, with different citation policies. What is documented, from primary sources where they exist:

ChatGPT — leans on the most-cited domain types. OpenAI has not published a description of the index behind ChatGPT’s browsing and search. The best public evidence is behavioral: Ahrefs’ monthly most-cited-domains study, September 2026 edition (updated September 2, 2026: https://ahrefs.com/blog/most-cited-domains-in-chatgpt/), finds citation share concentrated in community and peer sources (reddit.com at 16.8% mention share), review and reputation platforms, comparison media and official product/pricing catalogs. Our full analysis of what those types mean for B2B is in Reddit is ~17% of ChatGPT’s citations. The label: index composition is inferred from citation data (one vendor’s pipeline, one surface, snapshots), not from OpenAI documentation.

Gemini — grounded on Google Search, per Google’s own documentation. Google’s Gemini API documentation describes “Grounding with Google Search” as a first-class capability of the Gemini models: the model generates search queries, executes them against Google Search, and returns responses “grounded in the search results” with citations (Google AI for Developers, retrieved 2026-10-19: https://ai.google.dev/gemini-api/docs/generate-content/google-search). One scoping note: that is documentation of the model family’s grounding capability; Google does not publish a separate description of the consumer Gemini app’s retrieval path. The practical implication — your Google index footprint (crawlability, structured data, third-party corroboration) is what Gemini answers draw on — follows directly from the documented capability and is the clearest surface-to-asset link of the four.

Claude — its own retrieval, with a Brave-powered search backend (reported, not officially confirmed). Anthropic launched web search for Claude on March 20, 2025: Claude searches the web itself and returns “up-to-date, cited information in conversational responses” (Anthropic: https://claude.com/resources/articles/web-search). Anthropic does not name the search provider in product documentation. The evidence that it is Brave is strong but indirect: Brave Search was added to Anthropic’s subprocessor list the day before launch, and TechCrunch reported the finding as “appears to be using Brave” (TechCrunch, March 21, 2025: https://techcrunch.com/2025/03/21/anthropic-appears-to-be-using-brave-to-power-web-searches-for-its-claude-chatbot/). Label: Brave backend is evidence-based, not officially confirmed. Observable consequence: Claude’s answers track an index that is not Google’s, and Claude is visibly more selective about when it searches at all — many answers come from the model’s own knowledge with no citations. If you are cited by Google but not by Claude, that is a real, measurable gap, not noise.

Perplexity — the product is built around citations, density claim unverified. Perplexity describes its search product in its own words as delivering “accurate, cited answers from the web” (Perplexity product page, retrieved 2026-10-19: https://www.perplexity.ai/hub/products/search), and every answer displays its sources as numbered, clickable citations. It is, by its own design, the surface where a citation always comes with a visible link. The specific claim that Perplexity is the most citation-dense of the four surfaces per answer — several industry blogs assert it, typically citing 3–10 numbered sources per answer — is unverified: we have not found an official or independent measurement comparing per-answer citation density across the four surfaces, and vendor-adjacent blogs are not a neutral source. Treat it as a hypothesis until the harness data settles it.

The structural takeaway: one site, four different indexes, four different citation behaviors. A single-surface optimization program is, by construction, a partial one.

The sample run (labeled sample — illustrative, not live data)

Category: project management software. Shape of a run: ten category money queries × four surfaces, each answer captured verbatim and scored against the rubric in our methodology (cited with link / mentioned without link / absent, plus whether the answer included pricing, comparisons or review-platform evidence).

The table below is an illustrative sample: it shows the exact shape a four-surface run produces, with example values consistent with the documented retrieval mechanics — not a measurement of any real brand on any real date. It is not from the scheduled harness, it has no month-over-month delta, and it exists to show the shape of the data and the mechanics it demonstrates. When the harness completes its first live runs, this section is replaced with real data — the exact query set, per-surface scores and verbatim captures, all published — and the update log records the swap.

What a completed run in this shape looks like (sample values — illustrative, for one hypothetical brand across the 10-query set):

Query type (sample set of 10) ChatGPT Perplexity Gemini Claude
Category: “best project management software for [team type]” (3 queries) Cited 1/3 · mentioned 2/3 Cited 2/3 · mentioned 1/3 Cited 1/3 · mentioned 1/3 Cited 0/3 · mentioned 1/3
Comparison: “[X] vs [Y]” for two top brands (3 queries) Cited 2/3 Cited 3/3 Cited 2/3 Cited 1/3
Trust: “[brand] alternatives”, “[brand] pricing” (2 queries) Cited 1/2 Cited 2/2 Cited 1/2 Cited 0/2
Pain-point: “[category] for [specific workflow]” (2 queries) Cited 0/2 · mentioned 1/2 Cited 1/2 Cited 0/2 · mentioned 1/2 Cited 0/2
Surface total across the 10 queries 4 cited / 3 mentioned / 3 absent 8 cited / 2 mentioned / 0 absent 4 cited / 3 mentioned / 3 absent 1 cited / 2 mentioned / 7 absent

How to read this sample — and what it cannot show:

  • The spreads are the point, the counts are not. The example values mirror the documented mechanics — Perplexity’s citation-first product design, Gemini’s grounding on Google, Claude’s selectivity — but this is illustrative, not measured, and it is not a market measurement. It demonstrates that surface-level totals diverge for the same brand and the same queries, which is the finding a single-surface program misses entirely.
  • Citation ≠ the same citation. In the sample values, on “X vs Y” queries the surfaces name different brands; on pricing queries Perplexity leans on the vendor’s own pricing page (the official-catalog type) while ChatGPT leans on community threads. The source-type split mirrors what the ChatGPT citation data shows for that surface (what ChatGPT cites); the cross-surface version of that mapping is exactly what the live harness runs will test.
  • What it is not: not your category, not a ranking, not a trend, not reproducible from this page (the exact ten queries and raw captures will be published with the first live run).

What to check per surface (5 minutes each)

You do not need the harness to do a one-off check. Run the same 5–10 money queries by hand:

  1. ChatGPT (5 min): log in with an account that has web search on. Ask one category query and one comparison query; read the sources under each answer. Note: does the answer cite review platforms, community threads, or vendor pages? Is your brand cited, mentioned, or absent? Note the UTM behavior too — clicks from ChatGPT often arrive at your site without your tracking parameters, which shapes how you’ll see this traffic in GA4.
  2. Perplexity (5 min): no account needed for basic use. Run the same queries; count the numbered citations. Because every source is visible, this is the fastest surface to audit and to trace: open each cited source and ask why it was chosen over alternatives.
  3. Gemini (5 min): in the Gemini app, ask the category query and one pricing query. Check the cited sources for whether they track what Google Search returns for the same query — per Google’s documented grounding capability, that is the behavior to expect. If your Google results and Gemini’s citations diverge, that divergence is a finding in itself.
  4. Claude (5 min): enable web search in your profile settings, then run the queries. Two things to note: how often Claude searches at all (absence of citations is a signal, not a failure), and on the queries where it does cite, which sources — since the index is not Google’s, your “we rank on Google” evidence will not transfer.

Record everything in one table: query, surface, cited/mentioned/absent, sources named, one line on answer quality. Ten queries × four surfaces is an hour of work and tells you more about your AI visibility than any tool’s demo. The per-surface checklists and the 12-point site-side checklist are in our B2B SaaS citation checklist.

Which surface to prioritize by buyer type

There is no universal “the most important surface” — the honest answer is buyer-dependent, and the mechanics above are what make it so:

  • Buyers who shortlist in chat (founders, ICs, first-time evaluators): ChatGPT is the highest-prevalence chat surface — the published referral-share data puts it in a wide lead, as the referral traffic numbers detail — and its citations concentrate in the peer/review/catalog types. If your brand is absent from those source types for your category, this surface is where absence bites first.
  • Buyers who research deliberately (analysts, engineers, procurement): Perplexity is the deliberate-research surface; its visible citations make it the only surface where a buyer (or a competitor) can see exactly which sources shaped the answer. High effort to earn; unusually transparent once earned.
  • Buyers who live in Google (the default for most existing traffic): Gemini answers are grounded on Google Search per Google’s documentation — so your Google footprint, third-party corroboration and structured data are the levers. This is the surface where existing SEO work transfers most directly.
  • Buyers who evaluate rigor (security, data, regulated categories): Claude’s selectivity means a citation there carries weight — and an absence there means you do not exist in a research path that is, for some categories, the one that matters.

The practical sequencing: check all four once (the hour above), then prioritize the one or two surfaces where your buyers actually evaluate and you are absent — that combination, not raw traffic size, is where the first fixes go.

Update log

  • 2026-10-19 — launch version. Methodology and per-surface retrieval mechanics published, with a clearly labeled illustrative sample of the run shape for the project management software category. No live harness data yet: the harness’s first scheduled runs are pending. Every subsequent refresh — the sample-to-live swap and each monthly run after it — will appear here with its date, query set and surface set. Until the live swap happens, treat every table on this page as illustrative, not as a measurement.

Try the free quick check

If an hour of manual checking does not fit your week: we run your money queries across ChatGPT, Perplexity, Gemini and Claude and send back a snapshot scorecard — cited, mentioned, or absent, per surface, with what each answer actually said. No setup on your side, no meeting required. Get the free quick check.