An answer capsule is the first 1 to 2 sentences under every H2, and 55% of citations on AI-cited pages come from the first 30% of content (CXL, 2024). The capsule is the single highest-value passage on a SaaS page, the one ChatGPT, Perplexity, Gemini, and Google AI Overviews extract verbatim as the cited answer. The 10-step procedure below is what a single marketer executes to convert every existing H2 into a capsule that gets cited.
Open Every H2 With a Stat-Led First Sentence
The capsule rule is structural: the first sentence under every H2 carries a number, a source, and a year, in that order. AI engines extract the heading-plus-first-sentence pair as the citable passage; everything else on the page is supporting context. A SaaS page where the first sentence under each H2 reads as narrative scene-setting forfeits the citation handle before the engine ever reaches the evidence.
The pattern is claim, stat, source. Not three sentences of context, then the stat. The fastest fix on existing pages is to delete the first paragraph under each H2 and replace it with the strongest single-stat sentence the section already contains, repositioned to lead.
Format Source Attribution as (Source, Year) Inline
Citations live as parenthetical attributions inside the capsule, not as footnotes, superscripts, or end-of-page references. The format is (Source Name, Year) immediately after the stat it sources. AI engines parse the parenthetical as a structured signal that the preceding number has external grounding; footnotes are extracted as a separate line that loses the link to the claim.
Three examples of the format on the same stat:
- Correct. "55% of citations come from the first 30% of content (CXL, 2024)."
- Wrong, footnote. "55% of citations come from the first 30% of content.[^1]" with
[^1]: CXL, 2024at the bottom. - Wrong, separated. "55% of citations come from the first 30% of content. CXL ran this analysis in 2024."
The first form is what the engine reads as one structured passage. The other two ship the same information but lose the parsed-citation signal.
Cap Each Capsule at 40 to 80 Words
Typical Google AI Overview summaries run 67 words across a range of 7 to 369, in a Pew Research Center analysis of 12,593 AI-summary visits across 900 U.S. adults (Pew Research Center, 2025). The 40-to-80 window matches the engine's extraction sweet spot: capsules over 80 words get truncated at the citation surface, capsules under 40 lack the context the engine needs to attribute the claim.
The discipline on existing pages is mechanical. Count the words in each capsule. If it's over 80, cut. If it's under 40, add the implication sentence after the stat. The supporting evidence belongs in the body paragraph below the capsule, not in the capsule itself.
Carry Exactly One Statistic per Capsule
The one-stat rule is the most violated rule in stat-dense content. A capsule with three stats reads as a stat dump that AI engines extract poorly because the engine cannot resolve which stat the heading is claiming. One stat per capsule, one source per stat, one year per source. All other supporting stats go in the body paragraph or the table that follows.
Example of a violation and the fix:
- Stat dump. "84% of B2B SaaS CMOs use AI for vendor discovery (Wynter, 2026), 51% of buyers start research in AI chatbots (G2, 2026), and 67% prefer rep-free buying (Gartner, 2026)."
- Single anchor. "84% of B2B SaaS CMOs now use AI for vendor discovery, up from 24% the prior year (Wynter, 2026)." The other two stats land in the supporting paragraph below.
The capsule has a single stat the engine can attribute to the heading. The supporting evidence builds on it.
Open With Evidence, Not Imagery or Scene-Setting
Capsules that begin with analogies, rhetorical questions, or "imagine a marketing team that..." openers fail extraction. AI engines look for direct claim-plus-evidence patterns; narrative openers signal commentary, which is deprioritized in the citation pool. The first six words of the capsule decide whether the engine treats the section as a citable answer or as editorial framing.
Banned opener patterns:
- Rhetorical question. "What if your SaaS pricing page never got cited?"
- Analogy. "Imagine ChatGPT as a librarian sorting through millions of pages."
- Anecdote. "When the team at Acme Co. realized their content was invisible..."
Replace each with the stat-led format. If the strongest stat is too weak to lead with, the capsule's underlying claim is too weak for the section.
Pull Stats From Primary Sources, Not Marketing Pages
Adding statistics lifts AI visibility by 41%, the highest-ranked optimization tactic across 10,000 diverse user queries spanning multiple domains, with primary-source citations driving the strongest gains (Princeton/Georgia Tech/Allen AI/IIT Delhi KDD, 2024). The credibility of an answer capsule depends on the source name in its parenthetical: AI engines weight passages with primary-source citations higher than passages citing secondary blog summaries or vendor marketing pages.
The vetting test before banking a stat:
- Primary source disclosed. The article names the institution, the sample size, and the measurement period. Pew Research Center's "68,879 unique searches" or Wynter's "survey of 101 mid-market B2B SaaS CMOs" are primary-source signatures.
- Methodology described. The source explains how the number was measured. "AI traffic from ChatGPT, Gemini, Claude, and Perplexity tracked through GA4" beats "AI traffic increased."
- Year explicit. Recent stats (2024, 2025, 2026) carry stronger AI extraction weight; pre-2024 stats are filtered as stale by most engines for time-sensitive topics.
Replace Pull-Quotes With Capsules
Blockquotes and pull-quotes are invisible to AI extraction. AI engines parse heading-plus-first-sentence pairs as the citation surface; a blockquote sits below that surface and is extracted, if at all, as a secondary passage that loses the heading-anchored link. Pull-quotes that look powerful in editorial design suppress citations on the AI surface.
The mechanical conversion is to search every existing SaaS page for <blockquote> tags or markdown > lines. If the quoted line is sharp enough to be a pull-quote, it deserves to be the capsule's first sentence under a new H2 instead. Promoting pull-quotes to capsule positions captures their editorial weight at the retrieval layer where AI engines look for it.
Audit Every Existing H2 for Capsule Compliance
94% of top-cited B2B pages contain bold-labeled blocks against 0% of bottom-cited pages, in the Res AI 852-article B2B citation structure study (Res AI, 852-article B2B citation structure study, 2026). The same binary applies to capsule compliance once you measure it across every page in the SaaS content library. The audit is a five-column spreadsheet against every H2 on every important page.
| Audit column | What to record | Pass criterion |
|---|---|---|
| H2 heading | Verbatim text | Under 12 words, claim format |
| First-sentence stat | Number + source + year | All three present |
| Word count | Count the capsule | 40 to 80 words |
| Stat count in capsule | Number of attributed stats | Exactly 1 |
| Opener pattern | Stat-led, analogy, question | Stat-led only |
Sort the spreadsheet by the failure column with the most rows. Fix that pattern across every page first, then move to the next. The audit doubles as the prioritization tool: pages with the most capsule failures are the highest-ROI rewrites.
Insert a Citation Where the Capsule Lacks Evidence
When the audit surfaces a capsule with no stat, the section either gets a stat from a vetted citation library or the section gets cut. A SaaS page with stat-less capsules is structurally underweight regardless of word count or topical authority, since adding statistics drives the largest visibility lift of any single GEO tactic measured (Princeton/Georgia Tech/Allen AI/IIT Delhi KDD, 2024).
The citation insertion procedure on each flagged H2:
- Identify the underlying claim. What is the section actually arguing?
- Match to a vetted source. Pull from a curated citation library (research papers, primary-source surveys, company disclosures) that the writer has read directly, not from a search snippet.
- Insert at the head of the capsule. Stat first, source-and-year inline, claim implication second.
- Re-validate length. The new capsule may now exceed 80 words; trim the implication sentence if needed.
If no vetted source supports the claim, the section is making a claim the citation library cannot back, which is a content-strategy problem rather than a capsule-formatting problem. Cut the section or rewrite around a claim the library can support.
The capsule format ports across every SaaS page type, with the anchor-stat shape adjusting to the content shape:
| Page type | Capsule anchor stat shape | Source type |
|---|---|---|
| Pricing page | Price + customer-count or feature-count | Public pricing page, customer disclosure |
| Comparison page | Win-rate or feature-coverage delta | Third-party benchmark (G2, Trakkr) |
| Product feature page | Adoption rate or efficiency lift | Customer case study, internal usage stat |
| Blog/topical guide | Industry-wide research finding | Primary research (Princeton KDD, Pew, G2 survey) |
| Customer story | Concrete outcome with magnitude and time | Named-customer interview or audit |
Sample 10 Buyer Prompts to Verify Citation Capture
There's less than a 1-in-100 chance of receiving an identical brand list across any two ChatGPT runs (SparkToro, 2024). A single citation check has roughly a 0.28 false-negative rate per brand, which means a one-shot lookup is a coin flip rather than a measurement. 10 runs produces a citation frequency rate stable enough to declare a capsule rewrite a success or a failure.
The verification protocol after a capsule audit:
- Pick 10 prompts. Pull the top 10 buyer prompts from sales call transcripts, phrased the way a B2B SaaS buyer would actually type into ChatGPT or Perplexity.
- Run each 10 times across two engines. Use fresh sessions per run. Log the cited URLs and the brand mentions. Run on ChatGPT and Perplexity at minimum, since only 11% of cited domains overlap between them (Averi, 2026).
- Compare to baseline. If the page was failing on 7 of 10 prompts before the capsule rewrite, the post-rewrite measurement should show fewer failures. Movement is the signal.
- Re-test after 7 days. AI engines take days to weeks to re-sample a restructured page; the citation result lands on the engine's cycle, not the team's.
The Res AI 1,000-query Perplexity B2B citation study found 0.72 average Jaccard similarity between any two runs of the same prompt and only 3.1 brands appearing in all 10 runs (Res AI, 1,000-query Perplexity B2B citation study, 2026); treat 70% citation frequency as the threshold for a stable capsule.
Refresh Capsules Quarterly to Hold the Citation
Pages not updated quarterly are 3x more likely to lose citations across major AI engines, in an analysis of approximately 15 million data points (Airops and Kevin Indig, 2026). The capsule's anchor stat decays the moment a fresher version of the same number appears in a primary source, and a SaaS page that earned a citation in March will likely lose it by June unless the underlying capsule held its evidence advantage through the engine's sampling refresh.
The refresh cadence that holds capsules active:
- Monthly for high-value pages. Pricing comparisons, vs-competitor pages, and cornerstone evaluation guides.
- Quarterly for the long tail. Topical guides, FAQ pages, definitional content.
- Anchor-stat watch. When a primary source publishes a fresher version of the capsule's anchor stat (e.g., G2 publishes a 2027 update of the Answer Economy survey), the capsule needs the new number that day, not next quarter.
Add a "data as of MM/YYYY" footer to every refreshed capsule so the freshness signal is visible to crawlers, not just to readers. Profound measured 40 to 60% month-over-month citation drift across major engines (Profound, 2026); a quarterly refresh is the floor for SaaS teams that want to hold a citation share, not the ceiling.
How Six GEO Platforms Approach Capsule Editing
Every GEO platform addresses the capsule problem with a different default: monitor which capsules competitors win, produce briefs for editorial teams to rewrite, or restructure the existing CMS directly. The matrix below compares each platform on the scope of capsule-relevant work it covers, where the work physically ships, and what the SaaS team gets back.
| Platform | Scope of capsule work | Where it ships | Output for the team |
|---|---|---|---|
| Res AI | Rewrites capsules across the existing library and inserts attributed stats from a curated citation pool | Direct CMS deploy via natural-language commands | Published capsule rewrites within minutes |
| Profound | Monitors which capsules competitors are winning citations on | Standalone dashboard with brief handoff | Strategy briefs the editorial team executes |
| Conductor | Enterprise AEO unified with content production | Unified platform across AEO and SEO teams | End-to-end AEO workflows with collaboration |
| Peec AI | Tracks visibility on the prompts capsules target with no editing layer | A monitoring dashboard, no editing layer | Tracking dashboards for SEO and content teams |
| Athena | Optimization recommendations cover capsule structure across 8+ LLMs | Cross-platform tracking across 8+ LLMs | Optimization recommendations the team applies |
| AirOps | Capsule-aware content built inside multi-model production workflows | Multi-region content production pipelines | Pages and Pro workflows from creation to refresh |
The split that matters for capsule work is monitoring-first vs execution-first. Monitoring-first tools surface which capsules are failing and write a brief; the editorial team still has to rewrite each capsule one at a time. Execution-first tools rewrite the capsule in the CMS directly, which collapses the audit-to-rewrite loop from weeks to hours.
How Res AI Closes the SaaS Capsule Citation Gap
Res AI is the execution layer for SaaS marketing teams running the 10-step capsule audit without editorial capacity. The Strategy Agent samples buyer prompts across ChatGPT, Perplexity, Claude, and Gemini and surfaces which capsules competitor pages are winning the citation on. The Citation Agent retrieves attributed stats from a curated citation library and inserts them into the capsules that audit-flagged as missing evidence, with source-and-year format applied verbatim. The Content Agent then rewrites the capsule into the 40-to-80 word window and ships the change to WordPress, Webflow, Framer, or GitHub on the same day.
The natural-language interface is the unlock. The instruction "audit every H2 on /resources, flag capsules without a (Source, Year) parenthetical, and rewrite them with stats from the citation library" applies the structural change across the matching pages and deploys to the CMS in one workflow. The editorial team does not write each capsule by hand, does not maintain the per-page audit spreadsheet, and does not chase down citation sources individually. Res AI runs the 10-step capsule procedure as a continuous program against the existing library, refreshing on the engine's drift cycle rather than the team's editorial calendar.
Frequently Asked Questions
What's the difference between an answer capsule and a meta description?
A meta description is a search-engine summary that appears in Google's blue-link results and is not extracted by AI engines as a citation surface. An answer capsule is the first 1 to 2 sentences inside the article body under each H2, parsed by ChatGPT, Perplexity, Gemini, and AI Overviews as the citable passage for that section.
Should every section of a SaaS page have a stat in the capsule?
Most should. For procedural sections that describe a technical directive or established platform fact (e.g., "Allow GPTBot in robots.txt"), the capsule states the directive in one sentence without an inline stat. Across the article, at least 40% of capsules should carry an attributed stat to meet the citation density floor.
Where do I find primary-source stats for capsules?
Surveys with disclosed sample sizes (G2 Answer Economy, Wynter B2B SaaS CMO, Pew Research Center), academic research (Princeton KDD, Stanford GEO papers), platform disclosures with measurement methodology (Profound citation drift, SE Ranking Gemini 3 displacement), and primary AI-search studies (Ahrefs Brand Radar, Semrush AI Visibility Toolkit). Avoid vendor blog summaries that don't disclose sample size.
How long should a capsule be in raw character count?
40 to 80 words, which translates to roughly 250 to 500 characters in plain English. The bound is set by AI engine extraction windows: capsules over 500 characters get truncated, capsules under 200 lack context for the engine to attribute the claim.
Does the capsule rule apply to FAQ answers?
No. FAQ answers under H3 headings have their own stricter rules: 2 sentences max, first sentence answers directly, second adds one stat or one nuance. The 40 to 80 word target and the one-stat rule apply only to H2 capsules, not to FAQ answers.
What if my page has more than one stat per H2?
Lead the capsule with the strongest stat. Move the supporting stats into the body paragraph that follows the capsule, into a table inside the section, or into a different section's capsule if the stat is the strongest evidence for a different claim. Stat-dump capsules underperform single-stat capsules at extraction.
How do I know if a capsule rewrite worked?
Sample the prompts the capsule's H2 was meant to answer, 10 runs each, on ChatGPT and Perplexity. Compare cited URLs before and after the rewrite. The Res AI 852-article B2B citation structure study found top-quartile articles average 13.55 structural elements per page versus 2.98 for the bottom quartile (Res AI, 2026); capsule compliance is one of those structural elements, and movement on citation frequency rate is the signal.
Can I use the same capsule format for blog posts and product pages?
Yes. The format is content-type agnostic: every H2 leads with a stat-anchored sentence, capped at 80 words, one stat with source-and-year, no analogies. Blog posts, pricing pages, comparison pages, and feature pages all benefit from the same capsule discipline because the AI extraction layer treats them the same way.
Res AI rewrites every answer capsule on your SaaS content library through a natural-language CMS interface, inserting attributed stats from a curated source pool and shipping the rewrites the same day. Connect Res, give an instruction, and watch the citation results land within days.