Search “how to get cited in ChatGPT” and the results are generic: write well, add structured data, get on Wikipedia, be on Reddit. That advice is not wrong, but it is aimed at a consumer publisher, not a company selling software to businesses. The queries that actually matter to a B2B SaaS — comparisons, alternatives, pricing, trust — are a different population from the queries those guides optimize for. This post leads with the citation data that should shape what you do, then gives a twelve-check list written for a B2B SaaS site. And it makes no promises: nothing here guarantees a citation, because nothing short of the models themselves can.

What the citation data says

Two independent datasets are worth anchoring on.

Profound, ~700,000 ChatGPT conversations (October–December 2025). Profound analyzed approximately 700,000+ conversations from U.S.-based, English-language users on ChatGPT.com, in a study published February 3, 2026 (https://www.tryprofound.com/blog/chatgpt-citation-sources). Three findings matter here:

  • About 18% of conversations trigger at least one web search — and that rate held steady across all three months studied. Most ChatGPT conversations never look at the web at all.
  • Searches happen early, then drop off. Turn 1 is 2.5× more likely to produce citations than turn 10. Profound’s read: opening questions need factual grounding; follow-ups tend to be clarifications that don’t.
  • When ChatGPT searches, it doesn’t settle for one source — roughly 6 unique citations per cited conversation, and Profound notes the count was trending upward across the window.

Three operational consequences:

  1. You are competing for a slot in a cited answer, not ownership of one. With about six sources per conversation, being cited at all is the win; being the only citation is not how this works.
  2. The query that starts the research journey is the one to own. “Best [category] for [use case]” is turn-1 material. A follow-up deep dive is not where citations concentrate.
  3. “Get on Wikipedia and Reddit” is not a strategy — see next section.

Ahrefs most-cited domains (September 2026). Ahrefs tracks the domains ChatGPT cites across a broad set of U.S. queries, published monthly and updated September 2, 2026 (https://ahrefs.com/blog/most-cited-domains-in-chatgpt/). reddit.com leads with a 16.8% mention share — where mention share is a domain’s citations as a percentage of the summed citations of the top sources. The rest of the list reads like a consumer-research catalog: Wikipedia, Consumer Reports, Forbes, Healthline, Walmart, Target, Edmunds, TripAdvisor.

Notice what is not on the list: the G2s, the industry analysts, the vertical review sites that B2B buyers actually consult. That is a scope artifact — the study covers all topics — but it points somewhere real: general-purpose citation is dominated by consumer domains. B2B citation happens in a narrower, more competitive set of domains around each software category, and that set is where a SaaS company should be playing. If you want the full 50-domain table worked through — source types, what they reveal, and where the B2B gap sits — we did the data deep-dive in Reddit is ~17% of ChatGPT’s citations: what the most-cited domains mean for B2B SaaS. This post is the checklist; that one is the data read.

Why “get on Wikipedia + Reddit” fails a B2B SaaS

The Wikipedia-and-Reddit advice targets category queries: what is X, best X for me. For a consumer brand those are the money queries. For a B2B SaaS company they mostly are not.

  • Wikipedia does not cover vendors. It covers categories and concepts. A Wikipedia article on “customer relationship management software” will not name you, and trying to insert a product into an encyclopedia article is a losing fight against editors.
  • Reddit is not a controllable channel. There are Reddit threads where your product is compared with competitors — and some are favorable. But threads are scattered, stale, inconsistent in quality, and not something a B2B site can systematically optimize.
  • Your money queries are not category queries. They are: “[your product] vs [competitor]”, “best [category] for [your use case]”, “[competitor] alternatives”, “is [category] worth it for [use case]”, “[product] pricing”. A buyer asking these in ChatGPT is in the middle of a purchase decision, and the sources that answer them are comparison content, pricing pages, review platforms, analyst coverage — not encyclopedias.

There is one nuance from Profound worth keeping: sources are cited in clusters — a domain tends to appear alongside the same trusted set of domains, in each vertical. So the B2B version of “be near Wikipedia” is: become the source that appears alongside the review platforms, analysts and peer-reviewed write-ups your buyers already trust. That is checks 7 and 8 below.

The 12 checks

Each check: what, why, and a one-line how. Run them in roughly the order listed — cheap, high-leverage items first.

1. Structured data on the pages that get cited. What: JSON-LD (Organization, Product, FAQ, HowTo where it fits) on product, pricing, comparison and docs pages. Why: It is the machine-readable layer every AI-surface stack can consume, and its absence is a free fix you skip at your own risk. How: Add or repair JSON-LD on your top ten pages and validate it.

2. A dedicated page for each top competitor comparison. What: “[Your product] vs [competitor]” pages for the five competitors your buyers actually compare. Why: This is the highest-intent citation surface that exists for B2B SaaS, and answers to comparison queries need a page that is already shaped like an answer. How: One page per pair, real feature deltas and honest trade-offs — a page that only praises you reads as marketing to a judge.

3. A real pricing page, not a “contact sales” wall. What: Machine-readable pricing: tiers, numbers, at least the range, and what each tier includes. Why: “How much does X cost” and pricing comparisons are recurring query classes, and a wall produces an answer that cites someone else — a review site, a competitor with public pricing. How: Publish tier names and price points; “contact sales” can stay as an option, not the only answer.

4. Docs that answer, not document. What: Docs organized around questions (how to, how it works, what does it do) rather than feature dumps. Why: Profound’s turn-1 pattern means docs are most citable for the first factual question in a research journey — “how does X work” — which is exactly how a docs page should be written. How: Audit your top 20 doc pages against real first questions and retitle/rewrite the ones that fail.

5. An llms.txt that is actually valid. What: A spec-compliant llms.txt (+ markdown alternates for key pages), per the v2 spec. Why: It is one check, not the strategy — but it makes the machine-readable path cheap and reliable for the agents that fetch it. How: Run the 20-minute spec-based audit in our llms.txt post.

6. Freshness on the pages answers lean on. What: Dates on content, and a refresh cadence for the pages that answers lean on (category, comparison, pricing). Why: Answers favor current, specific content over evergreen generalities — and a 2023 comparison page is stale by construction. How: Put a review date on every money page and refresh the five most-cited ones first.

7. Third-party citations you can legitimately earn. What: Presence where B2B buyers source opinions: G2/Capterra-style review platforms, Product Hunt, analyst and vertical coverage, industry publications. Why: This is the B2B version of Profound’s clustering finding — you get cited alongside the domains already trusted in your category, and these are those domains for software. How: Fix the cheapest gap first: a complete, current profile where you are thin; honest responses where you have bad reviews.

8. Something original worth citing. What: Original data, benchmarks, or a position with real information density — not a listicle. Why: Models triangulate from multiple sources; an answer that needs a data point you alone have will cite the source that has it. How: One data-backed piece per quarter is a defensible cadence. No minimum, but zero is a gap.

9. Crawlability and speed. What: No bot blocking on the money pages, fast TTFB, clean server responses. Why: A page an AI system cannot fetch cleanly cannot be cited, however good it is. How: Fetch your top pages as a plain HTTP client (no JS), time them, and fix what is blocked or slow. When you do audit the blocking, use the LLM crawlers reference — one row per crawler, with the official doc behind each — so you are blocking training bots and allowing search bots deliberately, not by accident.

10. Stable URLs. What: No restructures that 404 the pages you want cited. Why: Cited URLs die quietly — a 301 from an old page loses the exact URL the answer was using. How: Before any restructure, list your money pages and verify each keeps resolving.

11. A trust page an answer can link to. What: Security, compliance, uptime, and customer references on real pages with real content. Why: “Is [category] worth it / safe for [use case]” queries land on trust content, and an AI answer needs a specific, citable page to point at. How: One page, concrete claims, named customers where possible.

12. A measurement baseline. What: Your money queries run across the AI surfaces, captured, and diffed — before you change anything else. Why: Without a baseline, “are we being cited?” is a vibes question, and you cannot tell which of the eleven checks above worked. How: This is the first thing we do in the methodology — query map, four surfaces, monthly diff.

What to measure after

Once the checks are in, measurement is the whole game — and it is easier than the tools’ dashboards make it look.

  • Citation share on your money queries, month over month. Cited, mentioned, or absent — per query, per surface, versus the competitors the answers name instead. That single number is defensible if it is reproducible from the query map and the raw captures.
  • The raw captures themselves. Verbatim answer excerpts are what let you (or your board) verify a score. No screenshot theatre.
  • A one-page monthly number. Citation share plus what changed and what you did — the shape is in the one-page report template.
  • Referral rows from AI surfaces in GA4 — real but small; treat them as a signal, not a KPI. The honest picture is in how to measure AI search traffic in 2026.

And one caution before you spend a quarter on this: citations and clicks are different metrics. The published CTR evidence on AI answers (the Ahrefs and Seer studies) is about Google surfaces and it matters, but “being cited” and “getting traffic” are not the same number. Measure citation presence first; traffic follows where it follows.

If you want the shape of the end state — your money queries across ChatGPT, Perplexity, Gemini and Claude, in one snapshot scorecard — the free quick check runs exactly that and sends the result. No meeting required to start.