When a buyer asks ChatGPT "What's a good CRM for a 10-person sales team under $50 per seat?" the AI fans the question into sub-queries, retrieves candidate pages, and synthesizes an answer that names specific vendors. Whether your SaaS gets named in that answer is a function of multiple signals — but a disproportionate amount of the weighting happens at your pricing page specifically. Pricing pages are the page AI engines visit when they need to verify a recommendation is real, current, and accurately positioned.
We wanted to understand exactly which pricing-page patterns correlated with strong AI citation outcomes. So we ran a structured audit of 50 B2B SaaS pricing pages across categories — CRM, project management, marketing automation, developer tools, design platforms, fintech, HR tech. Companies ranged from Series A (around $10M ARR) to public (Salesforce, HubSpot, Atlassian). For each pricing page, we measured 12 structural signals and tested 8 buyer prompts across ChatGPT and Perplexity to see which companies got cited and how.
This post is the result. The headline finding: one specific pricing-page pattern correlated with strong citation outcomes 4 times more reliably than any other signal we measured. Most SaaS pricing pages don't have it. Almost no template builders ship it by default.
The methodology, briefly
For each of the 50 SaaS companies, we audited their pricing page against the following structural signals:
- Explicit price numbers on-page (vs "Contact sales" / "Custom pricing")
- Annual vs monthly toggle present
- Seat-based vs flat pricing structure described
- Feature comparison table in markdown-extractable format
- Schema.org Offer markup present and valid
- FAQ schema present
- Use-case identifiers ("best for", "ideal for") in plain language
- Team-size guidance per tier
- Free tier presence
- Annual discount percentage stated explicitly
- "Compare to [competitor]" sections
- Trust signals (logos, testimonials, customer count) on the pricing page itself
Then we ran 8 standardized buyer prompts through ChatGPT (browsing mode) and Perplexity, recording which of the 50 companies appeared in each response and what specifically got cited.
Prompts ranged from broad ("What's the best CRM for a 10-person sales team") to specific ("CRM under $30 per user per month with email tracking and pipeline reporting"). Each prompt was run 5 times across one week to account for response variance.
The headline finding
The structural signal that correlated most strongly with citation outcomes was plain-text team-size guidance per pricing tier. Specifically: text like "Best for teams of 5-20" or "Ideal for companies with 50+ employees" appearing as readable copy next to each tier name, not buried in a feature table.
Of the 50 pricing pages in our sample, 12 had this pattern. Of those 12, 9 were cited in at least one ChatGPT or Perplexity response to our buyer prompts. That's a 75% citation rate.
Of the 38 pricing pages that did NOT have plain-text team-size guidance per tier, 7 were cited in at least one response. That's an 18% citation rate.
75% vs 18% is a 4.2x difference. No other signal we measured came close to that gap. Schema markup correlated at 2.1x. Explicit pricing correlated at 1.8x. FAQ presence correlated at 1.4x.
Why team-size guidance is so powerful
The reason is mechanical. When a buyer prompts ChatGPT with "What's a good CRM for a 10-person sales team," the AI's query fan-out includes searches like "CRM for 10 person team" and "CRM small business 5-20 users." Pages that contain phrases like "best for teams of 5-20" or "perfect for sales teams under 25 people" match those sub-queries directly.
Pages without that text don't match. The AI can't infer it from a generic feature list. Even sophisticated pricing tables that list "10 users included" don't trigger the same matching, because the buyer language is "for teams of 10," not "10 users included."
This is the broader principle of query fan-out at work. AI engines decompose buyer questions into sub-queries that often look quite different from the original question. Content that explicitly answers the decomposed sub-queries gets cited. Content that requires inference doesn't.
The other patterns that mattered
After team-size guidance, the next strongest correlations were:
Use-case identifiers ("Best for X", "Ideal for Y") per tier: 2.4x citation lift. Closely related to team-size guidance, but for non-team-size use cases (industry vertical, workflow stage, deployment model).
Valid schema.org Offer markup with priceCurrency, price, and availability fields populated: 2.1x citation lift. This is the only schema signal that moved the needle meaningfully. FAQ schema also helped (1.4x), but Article schema on the pricing page actually correlated negatively (probably because pages using Article schema were misclassified as blog posts by some retrieval systems).
Explicit price numbers on-page: 1.8x citation lift. "Contact sales" pricing crushes your AI citation rate. Of the 12 companies in our sample that used "Contact sales" exclusively (no published numbers), zero were cited in price-conditional buyer prompts. They occasionally appeared in unqualified prompts ("best CRM") but never when the buyer specified a budget.
Annual discount percentages stated as numbers: 1.6x lift. "Save 20%" or "$348/year (vs $30/month)" outperforms "Save with annual" by a meaningful margin. AI engines extract the percentage when it's written as a number.
Feature comparison tables in plain HTML (not JS-rendered): 1.5x lift. About 20% of the pricing pages in our sample used React-rendered comparison tables that didn't appear in the HTML response. Those tables didn't help AI citation at all — the AI never saw them.
The worst-performing pattern
The pricing-page pattern that correlated negatively with AI citation was multi-step pricing flows that hide numbers behind an interaction. Specifically: pages where you have to click "See pricing" or fill in a company size before any numbers appear.
Five companies in our sample used this pattern. None of them got cited in any price-conditional buyer prompt. Even when the company was a category leader with strong domain authority elsewhere, the AI couldn't extract pricing data from the page and so didn't surface them in budget-sensitive recommendations.
If your pricing page has an interaction wall in front of the numbers, you have effectively opted out of the entire price-conditional query class. That's a huge portion of high-intent buyer research in 2026.
What this means for your pricing page
If you do nothing else from this post, do these three things:
- Add a plain-text "Best for [team size/use case]" line under each pricing tier. One sentence per tier. Match the language buyers use ("teams of 5-15", "sales orgs with 25+ reps", "freelancers and solo consultants"). Not in a tooltip. Not in a feature table. As visible body copy on the page.
- Make sure your prices are server-rendered HTML, not lazy-loaded after JS execution. Test this by viewing source on your pricing page. If the dollar signs don't appear in the raw HTML, AI engines can't see them.
- Add valid Offer schema to your pricing tiers. Even basic markup with name, price, priceCurrency, and url is enough to lift you out of the "no offer markup" cohort. Skip Article schema on pricing pages — it confuses retrieval pipelines.
Those three changes are sub-half-day work for most engineering teams. Based on our dataset, they should move you from the 18% citation cohort to the 75% citation cohort for price-conditional buyer queries.
Limitations of this study
A few caveats worth naming. Our sample of 50 pricing pages is statistically small. Causation vs correlation is genuinely hard to establish here — companies with strong team-size guidance per tier may also be the kind of companies that take other parts of their marketing seriously, and the citation outcomes might reflect that underlying quality more than the team-size text specifically. We controlled for domain authority (using Ahrefs DR) and category as best we could, but the confound exists.
Citation outcomes also vary by prompt variant and over time. We ran each prompt 5 times across one week, but ChatGPT and Perplexity are not deterministic systems and a different sample week could shift our percentages by 10-15 points. The 4.2x gap we found is robust to that variance, but anyone replicating this should expect their specific numbers to be slightly different.
Finally, this study looked only at B2B SaaS. The same patterns may or may not transfer to D2C e-commerce, professional services, or local businesses. Our broader case study work suggests similar patterns hold in B2B-adjacent categories, but we haven't run the controlled audit in other niches yet.
Want the full dataset?
We've published the anonymized dataset (company size bucket, category, the 12 structural signals, citation outcomes per prompt) as a reference for anyone wanting to extend the analysis. Email research@reffed.ai and we'll send the CSV.
If you want to see how your own pricing page scores, run a free Reffed audit — the audit includes a per-page check of structured data, content depth, and citation-friendliness signals, including a dedicated check for the patterns this study identified. If you want the structured curriculum on building citation-friendly content end to end, Reffed Academy Quickstart covers pricing-page optimization plus 25 other lessons for $147 founding price.