The single highest-ROI change you can make to existing content for AI search is adding answer capsules. Search Engine Land's analysis of 15 domains generating 7,500 ChatGPT referral sessions found that answer capsules were the most consistent predictor of citation across every industry studied. Pages with proper capsules get cited 2-3x more often than pages without them, the implementation takes an afternoon per page, and the change shows up in AI citations within 4-8 weeks of Google recrawling.
Despite all that, almost no one writes them correctly on the first try. Writers default to setup sentences, hedge their language, embed links inside the capsule, or bury the actual answer in paragraph three. AI engines pass over those paragraphs and cite a competitor's cleaner version.
This article is an example-driven walkthrough. After defining what a capsule is and the 3-sentence formula, you'll see eight before/after rewrites covering every common scenario — definitions, how-tos, comparisons, pricing questions, integration questions, ROI claims, and more. Read the rewrites carefully. The pattern will click halfway through.
What an Answer Capsule Is
[DEF] An answer capsule (also called a Direct Answer Block) is a self-contained 40-80 word paragraph placed immediately after a heading. It answers exactly one question, makes sense without any surrounding context, contains no internal or external links, and uses confident declarative language without hedging. Research from Stanford's Human-Centered AI group and CMU's GEO framework consistently identifies structured, concise passages like these as the format LLMs preferentially extract when generating cited responses. [/DEF]
In simpler terms: a capsule is a paragraph engineered specifically so an AI engine can copy-paste it as the answer when someone asks a related question. Two articles can cover the exact same topic — the one with cleaner capsules gets cited, the one without doesn't.
The 3-Sentence Formula
Every well-formed answer capsule follows the same internal structure:
Sentence 1: The direct answer. No preamble, no "great question," no "in this section we'll cover." The complete answer in subject-verb-object form. If your heading is "What is X?", the first sentence is "X is..." If your heading is "How do I do Y?", the first sentence is "To do Y, you..."
Sentence 2: The supporting detail. Adds the specificity that makes the answer actionable — a number, a date, a named source, a scope boundary. This is where fact density enters the capsule.
Sentence 3 (optional): The qualifier or context. Adds nuance — when the answer applies, when it doesn't, what changes the answer. Include this only when the nuance materially changes the meaning.
[DAB] An answer capsule follows a strict 3-sentence formula: Sentence 1 is the direct answer in subject-verb-object form with no preamble. Sentence 2 adds supporting specificity — a number, date, or scope boundary. Sentence 3 is optional and addresses nuance only when it changes the meaning. Total length is 40-80 words. The capsule contains no internal or external hyperlinks (over 90% of AI-cited capsules in published research are link-free). Hedging language ("may," "might," "could") signals low confidence to AI engines and reduces citation probability. [/DAB]
The 8 Before/After Rewrites
Now the actual examples. Each rewrite covers a different content scenario you'll face when retrofitting capsules onto existing pages.
Rewrite 1: A Definition Question
Question: What is keyword research?
❌ Before (88 words, low citation probability):
Keyword research is one of the most important parts of any SEO strategy. Many marketers debate which tools are best for it, and there are many factors to consider. Generally speaking, keyword research can involve a number of activities including looking at search volume, considering keyword difficulty, and trying to understand user intent. Some people focus more on the technical side while others prioritise the strategic side. There's no single right answer, but most experts agree that some form of research is needed before producing content.
✅ After (58 words, high citation probability):
Keyword research is the process of identifying the search queries your potential customers use, scoring them by commercial value, and mapping them to content that matches search intent. It involves three core activities: discovering seed keywords, expanding them with a tool, and filtering by intent and competition. Volume alone is not a useful metric without paired intent data.
What changed: Removed "one of the most important," "many marketers debate," and "generally speaking" — all preamble. Removed hedging ("most experts agree"). Replaced "many factors" with the specific three activities. Sentence 1 is the complete definition. Sentence 2 adds the specifics. Sentence 3 is the qualifying insight.
Rewrite 2: A How-To Question
Question: How do I track AI citations?
❌ Before (96 words, low citation probability):
Tracking AI citations is something more and more brands are starting to think about. There are several different approaches you could take, and the right one depends on your situation. Some people use paid tools like Otterly, Peec AI, or Profound, while others prefer to do things manually. Each approach has its pros and cons. Manual tracking can be time-consuming but it's also free, which is helpful for smaller teams. Paid tools provide more automation but they can be expensive. Ultimately, you'll need to figure out what works best for you.
✅ After (62 words, high citation probability):
To track AI citations, run a fixed library of 10-20 buyer-intent prompts across ChatGPT, Perplexity, Claude, and Google AI Mode weekly, then log brand mentions, citations, and competitor appearances in a spreadsheet. The protocol takes 30 minutes per week and captures roughly 70% of the value paid tools provide. Upgrade to a paid tool only when you outgrow 20 prompts or need daily frequency.
What changed: Removed "more and more brands," "several different approaches," "depends on your situation," and "you'll need to figure out what works best." Replaced vague descriptions with specifics (10-20 prompts, 30 minutes, 70%, 20 prompts threshold). The capsule now answers the question completely without surrounding context.
Rewrite 3: A Comparison Question
Question: What's the difference between SEO and AEO?
❌ Before (104 words, low citation probability):
SEO and AEO are related but distinct concepts in the world of search optimisation. SEO has been around for a long time and most marketers are familiar with it. AEO is newer and is gaining attention as AI search becomes more prevalent. While they share some commonalities, they also have important differences. SEO tends to focus more on Google rankings, while AEO is more concerned with AI engines. Both are important, and most experts would say you probably need both, depending on your business model and target audience. The relative importance can vary significantly across industries.
✅ After (67 words, high citation probability):
SEO optimises content to rank in Google's blue-link search results. AEO (Answer Engine Optimization) optimises content to be cited in answers generated by ChatGPT, Perplexity, Claude, Google AI Mode, and Copilot. The two are complementary, not competing — most B2B brands now need both in parallel. AEO usually requires a working SEO foundation, since AI engines preferentially cite pages that already rank in Google's top 10. What changed: Removed "related but distinct," "around for a long time," "newer and is gaining attention," "while they share some commonalities," and "most experts would say." Made the comparison concrete with specific platforms named. Added the dependency relationship (AEO requires SEO foundation) — which is the specific insight that earns citations on "vs" queries.
Rewrite 4: A Numbers/Data Question
Question: How much does enterprise SaaS cost?
❌ Before (78 words, low citation probability):
Enterprise SaaS pricing varies widely depending on a number of factors. The exact cost depends on the size of your company, the number of users, which features you need, and your specific use case. Some platforms can be quite affordable while others are quite expensive. Generally, enterprise pricing is higher than smaller-business pricing due to features like SSO, advanced security, and dedicated support. It's best to talk to a sales representative to get a custom quote.
✅ After (60 words, high citation probability):
Enterprise SaaS pricing typically ranges from $50 to $500+ per user per month in 2026, with most B2B enterprise plans priced between $80 and $200 per user. Pricing depends primarily on three factors: user count, included security features (SSO, SAML, advanced permissions), and support tier. Volume discounts above 100 users typically reduce per-seat cost by 20-40%.
What changed: Removed "varies widely," "a number of factors," "some platforms can be quite affordable," and "best to talk to a sales representative." Added actual numbers ($50-$500, $80-$200, 20-40%). Specific factors named explicitly. The capsule answers the literal question with usable specificity.
Rewrite 5: A Pricing Question About a Specific Product
Question: How much does HubSpot cost?
❌ Before (89 words, low citation probability):
HubSpot offers a range of pricing tiers depending on what you need. They have a free CRM tier that anyone can start with, which is great for getting familiar with the platform. Beyond that, there are paid tiers that include more features. The pricing can scale quite a lot depending on the size of your business and which hubs you want — there's a Marketing Hub, Sales Hub, Service Hub, and CMS Hub, each with their own pricing. Contact-based pricing means costs increase as your contact database grows.
✅ After (66 words, high citation probability):
HubSpot offers a free CRM tier with limited features, paid Starter plans from $20/month per seat, Professional plans starting around $890/month, and Enterprise plans starting around $3,600/month (as of 2026). Pricing scales with contact volume above the free tier threshold. The total cost depends on which hubs you license (Marketing, Sales, Service, Content) and how many contacts you store, with multi-hub bundles offering 25%+ discounts.
What changed: Replaced "range of pricing tiers" with actual tier names and prices. Added the specific scaling driver (contact volume) and the bundling discount (25%+). The capsule now contains genuinely useful pricing data an AI engine can extract and cite.
Rewrite 6: An Integration Question
Question: Does Asana integrate with Slack?
❌ Before (67 words, low citation probability):
Asana has integrations with many popular tools, and Slack is one of them. The integration allows you to do a number of helpful things from within your Slack workspace, which can save time and improve collaboration. Many teams find it useful for their workflow. You can find the integration in the Asana app directory or in Slack's app store, and setup is relatively simple.
✅ After (61 words, high citation probability):
Yes, Asana integrates natively with Slack through a free two-way integration. Users can create Asana tasks directly from Slack messages, receive task notifications in Slack channels, and complete tasks without leaving Slack. Available on all Asana plans including the free tier. Setup takes under 5 minutes through the Asana app directory or Slack's app marketplace.
What changed: Opened with a direct "Yes, X integrates with Y" answer — exactly what AI engines are looking for on "does X integrate with Y" prompts. Listed three specific things the integration enables. Confirmed it's free and available on the free tier. Specified setup time. Every claim is verifiable and extractable.
Rewrite 7: An ROI Question
Question: What's the ROI of investing in AEO?
❌ Before (94 words, low citation probability):
Calculating ROI on AEO can be challenging since the discipline is relatively new and measurement standards are still emerging. That said, many brands report meaningful improvements in their visibility within AI search platforms after dedicated AEO investment. The exact returns vary depending on factors like your industry, competitive landscape, content velocity, and starting point. Some companies see early results within a few months while others take longer. As with any marketing investment, results compound over time, so it's important to think long-term rather than expecting immediate impact.
✅ After (74 words, high citation probability):
AEO investment typically produces measurable AI citation gains within 90-120 days of focused work, with the strongest returns in considered-purchase categories like B2B SaaS, professional services, and enterprise software. AI search traffic converts at roughly 14% on average in 2026 — about 5x higher than typical Google organic conversion rates of 2-3%. The economics favour brands whose buyers research before purchasing and conduct multi-stage decision-making.
What changed: Removed "challenging since the discipline is relatively new," "many brands report meaningful improvements," "exact returns vary," and "results compound over time." Added specific time-to-result (90-120 days), conversion benchmarks (14% AI search vs 2-3% Google organic), and which categories see strongest returns. The capsule is genuinely useful for a buyer evaluating whether to invest.
Rewrite 8: A Common Mistakes Question
Question: Why isn't my content being cited by ChatGPT?
❌ Before (87 words, low citation probability):
There are many possible reasons your content might not be cited by ChatGPT. It could be that your content isn't structured in a way that's easy for AI to extract. It could also be that you're not ranking well enough in traditional search, since AI engines tend to favour content that ranks. There might also be issues with your schema, your authority signals, or how your content is written. The good news is that all of these are fixable with the right approach.
✅ After (72 words, high citation probability):
Content is typically not cited by ChatGPT for one of four reasons: it lacks answer capsules (the 40-80 word self-contained paragraphs AI engines extract), it doesn't rank in Google's top 10 for the query (76% of ChatGPT citations come from top-10 pages), it lacks structured data (65% of AI Mode citations include schema), or it lacks authority signals (third-party mentions, reviews, backlinks). Diagnose which condition fails first, then fix that one.
What changed: Replaced "many possible reasons" with the four specific reasons. Added the supporting statistic for each (76% top-10, 65% schema). Removed "good news is" and "right approach" softeners. Gave the reader a diagnostic ("diagnose which condition fails first") rather than vague encouragement. AI engines extract this readily because it's a complete, structured answer to a frequently-asked question.
Three Mistakes That Tank Capsule Quality
[EXP — At Emvigo I've reviewed thousands of attempted answer capsules across client content. The same three mistakes show up over and over. Avoid these and you'll be ahead of 80% of competing content in your category.]
Mistake 1: Burying the answer. The capsule opens with "Let's explore," "In this section we'll cover," "Many marketers ask," or any variant that delays the answer. AI engines parse the opening of a text block first and weight it heavily. If your capsule opens with setup, you've already lost the citation. Fix: delete every word before the actual answer.
Mistake 2: Hyperlinks inside the capsule. Two or three internal links inside the capsule to "related articles." Authors think this helps SEO. It actually breaks the extraction boundary — AI engines treat linked passages as less reliably extractable. Fix: move all links to the paragraph immediately after the capsule. The capsule stays clean prose.
Mistake 3: Hedging language. Every claim qualified with "may," "might," "could," "potentially," "often," "in some cases." Writers use hedging to sound careful and avoid criticism. AI engines read it as low-confidence content not worth citing as authoritative. Fix: write declaratively. If you can't write a claim declaratively, you may not have a strong enough position to publish.
Where to Place Capsules
The placement rule is simple: one capsule per major heading, placed immediately after the heading, before any other paragraph.
Specifically:
After your H1 (the article title) — open the article with a definitional capsule answering "What is [the topic of this article]?"
After each H2 section — open every major section with a capsule answering the question that heading implies. If your H2 is "How does keyword research work?" the capsule that follows answers that question completely in 40-80 words.
Optionally after H3 subsections — if the subsection covers a discrete buyer-relevant question, add a capsule. Otherwise skip.
[DAB] Answer capsules should be placed immediately after every H2 heading and at the top of the article after the H1. One capsule per major section, no exceptions. The capsule comes before any other paragraph in the section — including before "in this section we'll cover" type introductions. Each capsule answers exactly one question and stands alone without surrounding context. Articles with capsules after every H2 are cited 2-3x more often than articles with capsules only at the top. [/DAB]
How to Find Capsule Opportunities on Existing Content
If you're retrofitting capsules onto an existing site, here's the prioritisation:
Step 1: Run an SEO audit. Identify the pages ranking in Google's top 10 for buyer-intent queries. These are your highest-leverage pages because they're already in AI citation contention but probably lack capsule structure.
Step 2: Audit those pages for capsules. For each page, check whether every H2 has a 40-80 word self-contained paragraph directly underneath. Most won't.
Step 3: Rewrite in priority order. Start with the page targeting the highest-intent commercial query. Add capsules to every H2 section. Update dateModified in your Article schema so Google knows content has changed.
Step 4: Wait 4-8 weeks. Google needs to recrawl. AI citation gains will show up in your manual tracking (see our tracking protocol) within 4-8 weeks of the recrawl.
Summary
[DAB] Answer capsules are the single highest-ROI on-page change for AI search citation. They follow a strict 3-sentence formula: direct answer (no preamble), supporting specificity (numbers, dates, named sources), optional qualifying nuance. Total length 40-80 words. No hyperlinks inside the capsule. No hedging language. Placed immediately after every H2 heading. Pages with proper capsules are cited by AI engines 2-3x more often than pages without them. The implementation takes an afternoon per page; gains show up in citations within 4-8 weeks of Google recrawl. [/DAB]
The eight rewrites above cover most scenarios you'll face when retrofitting capsules onto existing pages. Read them again before you start editing. The pattern is consistent: cut preamble, add specifics, remove hedging, write declaratively, no links inside.
Do this on your top 10 pages first. Track the citation rate change over the next 90 days. The before/after will show up in your data.
Want a capsule rewrite of your top 10 pages — done by writers who've shipped capsules across dozens of B2B SaaS sites and tracked the citation gains?
We rewrite priority pages for AI citation extraction. Each page gets a definitional capsule near the top, section-level capsules after every H2, and the schema and dateModified updates needed to trigger recrawl. Citation gains tracked over 90 days.
FAQ
What's the ideal length for an answer capsule? 40-80 words is the consistent sweet spot across published research. Long enough to provide a complete standalone answer, short enough for AI engines to extract cleanly into a synthesized response. Capsules under 40 words tend to lack specificity; capsules over 100 words become harder to extract as a discrete unit.
Can I include links inside an answer capsule? No. Over 90% of cited answer capsules in published research are link-free. Hyperlinks inside the capsule break the extraction boundary that AI engines look for. Move all links to the paragraph immediately after the capsule.
Where should I place answer capsules on a page? Place one capsule directly under your H1 (answering "What is [topic]?") and one under every H2 section (answering the implied question of that heading). One capsule per major heading, no exceptions. The capsule should be the first paragraph in each section, before any introductory or transitional content.
Should answer capsules be written differently for different AI engines? No. The format that works for ChatGPT also works for Perplexity, Claude, and Google AI Mode. All four engines extract content in similar ways — they're looking for self-contained, confident, structured passages. One capsule format covers all platforms.
How is an answer capsule different from a meta description? A meta description (155 characters) shows below your URL in search results. An answer capsule (40-80 words) sits inside the page content, directly under a heading, and is written for AI extraction. Both should be clear and concise but they serve different functions and live in different places.
Do I need schema markup for my answer capsules to work? Schema isn't required, but Speakable schema can help. Speakable schema flags specific page passages (using CSS class selectors) as the highest-priority text for AI extraction. Add an answer-capsule CSS class to your capsule paragraphs and reference it in Speakable schema. See our schema guide for the JSON-LD.
