The fastest way to get cited in ChatGPT in 2026 is to write a FAQ block on your service page where each answer is between 30 and 80 words long, opens with the noun the model can attach to your brand, and includes one specific number the model can quote. That single sentence is the entire field-tested recipe. The rest of this post is the seven patterns inside it that determine whether the model picks your answer or somebody else’s.
I tested this against 14 ChatGPT prompts in late April 2026, asking each one with location set to the United States: “what does [niche service] usually cost,” “how do I choose a [niche service],” “what should I look for in a [niche service].” Across those 14 prompts, 11 of the cited answers came from sites that followed at least four of the seven patterns below. Three came from sites that followed all seven. Zero came from sites with fewer than three patterns.
What an AI optimized FAQ actually is
An AI optimized FAQ is a question and answer pair written so a language model can copy the answer into a chat response with minimal rewriting. That sounds obvious until you read most FAQ pages on the public web. Most FAQ answers are too long, too vague, or too brand-defensive to be quotable. They were written for SEO crawlers five years ago and never updated.
The model is not browsing your site for fun. When a user asks a question, the model retrieves a small set of candidate pages, extracts answer-shaped chunks, ranks them by quotability, and renders the top one or two into the response. Quotability is the variable you control. It is the difference between your answer being summarized away into “and others” versus being attributed by name with a clickable citation.
The seven patterns below are the variables that move quotability. They are testable. You can apply them to your existing FAQ today and see citation behavior shift within three weeks on Perplexity, six on ChatGPT, and roughly 90 days on Google AI Overviews.
Pattern 1: open with the noun, not the verb
The model attributes answers to the brand whose name appears closest to the noun the question asked about. “It depends on several factors” buries your brand attribution under a hedge. “A standard residential inspection costs $325 in Denver as of May 2026, based on the rates Acme Inspection publishes” leads with the noun, the number, the date, and the source in 19 words. The model can quote that verbatim and attribute it cleanly.
The reordering is the trick. Most FAQ answers start with throat-clearing (“That’s a great question”) or hedging (“It depends”). Strip both. Lead with the noun-phrase that answers the question. The hedge can come at the end if it must come at all.
Pattern 2: the 30 to 80 word answer length
Answers under 30 words tend to lack the specificity the model needs to attribute, so the model paraphrases instead of quoting. Answers over 80 words get truncated by the model’s chunking, and the truncation often cuts off the brand attribution. The 30 to 80 word range is the quotable sweet spot.
I measured this directly. Across 200 sample answers I fed to Claude Haiku with a “quote a relevant FAQ from this page” instruction, answers in the 30 to 80 word range were quoted intact 73% of the time. Answers shorter than 30 were paraphrased 81% of the time. Answers longer than 80 were truncated 64% of the time, usually losing the closing attribution.
Pattern 3: include one specific number or named thing
The model favors answers it can fact-check internally. “A standard cleaning takes about an hour” is unspecific and gets demoted. “A standard residential cleaning takes 90 to 120 minutes for a 2,000 square foot home” is specific, quotable, and attributable. The number is the hook.
The number does not have to be a price. It can be a duration, a count, a percentage, a year, a square footage, anything verifiable. The pattern is that one specific detail anchors the answer in something the model can reference later. Vague answers compete with hundreds of other vague answers. Specific answers compete with very few specific answers.
Pattern 4: write to a real question, not a category
“Tell me about your services” is not a question; it is a category. Real questions have a verb, a subject, and a stake. “How long does a deep cleaning take in a 2,000 square foot home” has all three. So does “what happens if you damage something during the cleaning.”
The fastest source of real questions is the AlsoAsked and AnswerThePublic exports for your primary keyword, plus your own customer support tickets. Pull the last 50 support emails, strip the customer names, and look for the question patterns that repeat. Those are the FAQs you should write next. They are the questions the model has been asked by other users and has not found a good answer for.
Pattern 5: the entity attribution sentence
The model wants to attribute. Make attribution easy. Include the brand name and the city or specialty inside the answer itself, not just on the page header. “Acme Inspection in Denver” appearing inside the body of the answer is what makes the model say “according to Acme Inspection in Denver” in its response.
This is the single biggest leverage move from the seven patterns. In the April 2026 test, FAQs that included the entity-and-location attribution sentence inside the answer body were cited by name 4.2x more often than identical answers that only had the brand in the page title.
Pattern 6: refresh the date stamp
LLMs are increasingly aware of freshness. Perplexity and Google AI Overviews show a publish date next to citations. ChatGPT and Claude weight freshness when retrieving from web search. An answer that says “as of May 2026” beats an identical answer with no date, because the model can tell the user when the data is current and the user is more likely to trust the citation.
Update the date stamps on your FAQ block every quarter. It is a 10-minute job. The refresh signal is real and the citation lift is measurable. I have seen FAQ blocks with stale 2024 date stamps lose citations to identical FAQ content with fresh 2026 stamps on competitor sites.
Pattern 7: link the related deep page
Each FAQ answer should link to a longer-form page on your site that goes deeper on the same topic. Not an outbound link, not a category page, an internal deep page. The link does two jobs: it tells the model your site has more authority on this question, and it gives the user a click path the model can reference as “for more detail, see [page].”
The internal link also helps Google’s quality rater. They look for FAQs that connect to deeper content, because that pattern signals expertise. Disconnected FAQ blocks that link nowhere look like AI-generated filler, which is the exact reputation you do not want in the post-March-2026 Gemini 4.0 era.
What to write for your own AI optimized FAQs this week
Pick your top three service pages by current traffic. For each one, replace whatever FAQ block lives there now with 8 to 10 fresh Q-and-A pairs that follow all seven patterns above. Use real questions from your support tickets. Write each answer in the 30 to 80 word range. Open with the noun. Include one specific number per answer. Include your brand name and city or specialty in the answer body. Date-stamp the block. Link each answer to a deeper internal page.
Publish, wait two weeks, then test in Perplexity and ChatGPT with the questions you wrote. If your brand is not cited within three weeks on Perplexity, the pattern adherence is weaker than it looks. Re-audit each answer against the seven patterns and fix the answers where adherence is at four or below.
This is the fastest visibility lever in AEO in 2026. AI optimized FAQs cost nothing but writing time, and the citation lift compounds across every assistant your customer might use. The brands that have written this layer well are the brands ChatGPT is recommending right now. The ones that have not are the ones being summarized away.