Skip to content
How-to guide

How to Write Content That AI Search Engines Quote

The short answer

AI engines quote passages that answer one question fully and still make sense on their own. Open every section with the answer in one or two plain sentences, name the subject instead of writing 'it', and add a sourced number or a short attributed quote. In one academic test, quotations, statistics and cited sources raised visibility, while keyword stuffing did not 3.

  • AI engines retrieve and rank passages, so write each paragraph to be quoted alone, with the answer in the first sentence.
  • Replace vague words with a named subject, a number, a date and a source. Never invent any of the three.
  • In a simulated test, adding quotations, statistics and citations beat keyword stuffing, which scored below the baseline on the main measure 3.
  • Ranking in the top 10 is not required to be cited. One passage that matches a sub-question can be enough 4.
  • Test with real prompts and keep a log. Nobody can promise a citation, and results vary between runs.

What do AI engines actually quote?

They quote passages, not whole pages. Perplexity says it breaks documents into self-contained spans, then retrieves and ranks those spans one by one 6. Google's ranking systems include passage ranking, which scores individual sections of a page as well as the page itself 7.

So the unit you write for is the paragraph. A quotable paragraph answers one question, names its subject and survives being copied alone into a chat answer.

Google says you need no special markup or 'AI files' to appear in AI Overviews or AI Mode. The page must be indexed and eligible to show with a snippet 1. That statement covers Google only. Other assistants set their own rules, but writing is the part you control in all of them.

Do it step by step

0 of 9 done0%

  1. Choose a page that already gets visits or sits near a sale. Write the three to five questions it should answer, in the words customers use.

    Tip: Take the wording from sales calls and support emails, not from a keyword tool.

  2. In 40 to 75 words, answer the main question so the paragraph makes sense alone. Name the subject. Include one number or fact with its source.

  3. Rewrite headings like 'Overview' into the question a customer asks. Open each section with the direct answer, then explain. One idea per section.

  4. Run the pronoun search from the templates. For every hit, name the subject. Replace 'many', 'leading' and 'often' with a number, a date or a deletion.

    Tip: If you cannot replace a vague word with a fact, delete the sentence.

  5. For each statistic, add the publisher, the year and a link in the same sentence. Open the source and confirm the number says what you claim.

    Tip: Write the year next to the figure. Old data without a year misleads readers and engines.

  6. Ask a real person on your team or a customer for a one-sentence quote and written permission. Print the name and role beside it. Skip this step if you have no genuine quote.

  7. Link the author name to an author page. Show a visible 'last updated' date that you change only when content changes. Say plainly if AI helped draft the page 2.

  8. Copy three random paragraphs into a blank document. Ask: does a stranger know the subject, the claim and the source? Fix any that fail. Then run the free Answer Block Checker.

  9. Run your prompt list before the edit goes live and write down the result. Re-test two and four weeks later, three runs per prompt, and keep the log.

    Tip: Change one page at a time at the start, so you can see which edit mattered.

Your ticks are saved in this browser only.

Copy-ready templates

Quotable section skeleton

markdown
## {Question a customer would ask}?

{Direct answer in one or two sentences. Name the subject. Include one fact.} [Source: {publisher}, {year}]({url})

{Why or how, in two or three short sentences. One idea only.}

> "{One-sentence quote}" - {Full name}, {role}, {organisation}

*Last checked: {YYYY-MM-DD}.*

Replace every brace. Delete the quote line if you have no genuine, approved quote.

Pronoun-opener search (regex)

regex
^(It|This|That|These|Those|They|There)\b

Paste into the find box of a code editor such as VS Code with regex mode on. Each match is a paragraph that may lose its subject when quoted alone.

Pre-publish checklist

text
[ ] Heading is a real question
[ ] First sentence answers it
[ ] Subject is named (no 'it' or 'this' opener)
[ ] One idea per paragraph, four sentences or fewer
[ ] Every number has publisher, year and link
[ ] Any quote has a real name, role and permission
[ ] Byline links to an author page
[ ] Visible last-updated date matches a real change
[ ] Facts are in HTML text, not only in images or PDFs
[ ] Page is indexed and not blocked to AI crawlers

Run it on every page before publishing.

Prompt test log

yaml
- prompt: "{question a customer would ask}"
  engine: "ChatGPT | Perplexity | Google AI Mode"
  date: "YYYY-MM-DD"
  run: 1
  named: yes_or_no
  linked: yes_or_no
  cited_urls: ["{url}"]
  notes: "{what changed since last run}"

Copy one block per prompt, per engine, per run. Three runs per prompt smooths out random variation.

What do you need, and how long does it take?

You need one existing page that matters to your business, a plain text editor and a short list of the questions that page should answer. Plan on roughly 90 minutes for your first page and half an hour for each one after. Those times are our own estimates, not measured data.

The free Answer Block Checker tests whether your opening answer is the right length and stands alone. The pattern search in the templates below finds paragraphs that start with a pronoun.

  • A page that is already indexed. If AI crawlers cannot reach it, fix that first with this crawl check.
  • Three to five real questions from customers, sales calls or support emails.
  • Facts you can prove: prices, dates, numbers, names, with a link for each.
  • Someone who can approve quotes and claims before you publish.

Does the way you write change whether AI quotes you?

Yes, in the one controlled test we can point to, and with limits. The GEO paper (Princeton, IIT Delhi and independent authors) tried nine content changes on 10,000 queries across 25 domains. The best change, adding quotations, scored 27.2 against a 19.3 baseline on its visibility measure, about 41% higher 3. Adding statistics gave about 31% and citing sources about 28% 3.

Read those numbers with care. The test used a simplified engine: it fetched the top five Google results and let gpt-3.5-turbo write the answer 3. The score measures how much of the answer draws on your text. It is not clicks, leads or revenue, and real assistants behave differently.

A second reason matters more for small sites. Ahrefs studied 863,000 keyword results and about 4 million AI Overview URLs. Fewer than four in ten cited URLs ranked in the first 10 blocks of the results page 4. Ahrefs suggests a likely reason is query fan-out, where an engine splits a question into several smaller searches and cites pages that rank for those sub-queries. It presents this as an explanation, not proof 4.

37.9%of URLs cited in Google AI Overviews also appeared in the first 10 blocks of the results page, which include ads, featured snippets and other results-page elements (Ahrefs, March 2026, 863,000 keyword results)4

What does a quotable paragraph look like?

A quotable paragraph puts the answer first, uses one idea, and holds one checkable fact. It runs to four sentences at most. The subject has a name, so the sentence works when pulled out of the page.

The example below is invented to show the change. Acme Billing is a fictional company, and its prices are not real.

Use the table to spot the habits that make passages hard to quote. Each row pairs the habit with the fix. If you want the deeper theory, read what an answer block is and what atomic content means.

Before and after (fictional company)
  • BEFORE: Many of our customers ask about this. It depends on a lot of factors, and our team is happy to help you find the right option.
  • AFTER: Acme Billing costs $49 a month for up to 5 users. Extra users cost $8 each. There is no setup fee. Prices last checked on 20 October 2026.
Seven habits that make a passage hard to quote. Reasons are our reasoning from how passage retrieval works, not measured results.
HabitLikely problemFix
Starting with 'It' or 'This'The passage loses its subject when copied aloneName the product, person or thing in the first sentence
Answer buried after a long introThe retrieved span holds the warm-up, not the answerPut the answer in the first sentence, then explain
Words like 'many', 'leading', 'often'Nothing to check or quoteState the number, the date and the source
Three ideas in one paragraphMatches none of the sub-questions wellSplit into one paragraph per idea
Headings like 'Overview'Gives no clue what question the section answersWrite the heading as the question a customer asks
Facts only inside images or PDFsText may never reach the engine's indexRepeat the facts as plain HTML text
Undated prices and policiesEngine cannot tell if the fact is currentAdd a 'last checked' date next to the fact

How should you use numbers, quotes and sources?

Add a specific wherever a vague word sits, and attach its source in the same sentence. The table below shows which changes scored higher in the paper's test, scored against an unchanged baseline of 19.3 3.

Keyword stuffing means repeating the target phrase. On the paper's main measure, position-adjusted word count, it scored 17.7 against a 19.3 baseline, about 8% lower. On its subjective-impression rating it did not beat the other methods, scoring about 20.2 3. Write the way people ask the question and stop there.

The authors also broke results down by how well a site already ranked. Adding citations lowered the score by about 30% for first-ranked sites and raised it by about 115% for fifth-ranked sites 3. Take this as a hint, not a rule: smaller, lower-ranked sites may have more to gain from showing their evidence.

We have not yet run a controlled test of these edits on our own pages; the evidence above is third-party. Never invent a quote, a statistic or a credential to chase these gains. Google treats false credentials and fake author details as deception 2. A quote needs a real named person, their role and their permission.

GEO paper: position-adjusted word count by content change (baseline 19.3). Source: [s3]
Content changeScoreChange vs baseline
Quotation addition27.2about +41%
Statistics addition25.2about +31%
Fluency optimisation24.7about +28%
Cite sources24.6about +28%
Keyword stuffing17.7about -8%

Do dates and author names matter?

Both help, but neither replaces substance. Ahrefs analysed 16.975 million cited URLs. URLs cited by AI assistants averaged about 1,064 days since first publication, against about 1,430 for organic results, with ChatGPT showing the strongest lean to newer pages 5. Cited pages were still about 2.9 years old on average, so old pages do get quoted 5.

Update a page when a fact changes, and show the date. Do not change the date without changing the content. Ahrefs notes that Google's John Mueller has warned against that trick 5. For old posts, follow the steps in how to refresh old content for AI search.

Google's people-first guidance asks three questions: who wrote this, how was it made, and why does it exist. It wants a byline linked to author background, a plain note on any automation or AI use, and original information beyond what other sources already say 2. Build the author page once, using this author page guide, and link every byline to it.

What mistakes stop good content from being quoted?

Most failures come from habits, not from missing tricks. These five are common writing habits that work against passage-level retrieval. They are our editorial judgement, not measured results.

  • Writing the answer last. Engines and readers both want it first.
  • Padding the opening with a definition of the industry. Cut to the claim.
  • Stacking statistics with no source. An unsourced number is easy to skip and risky to quote.
  • Copying other pages in new words. Google asks whether you add original information, reporting or analysis 2.
  • Hiding facts behind tabs, images or login walls that crawlers may not read.

How do you check it worked, and when is it not worth doing?

Test with real prompts. Pick five to ten questions a customer would ask, run each in ChatGPT, Perplexity and Google AI Mode before and after your edit, and log whether you are named or linked. Repeat each prompt three times, because answers change from run to run. Use the log template below, and see how to build a prompt list.

In Google's own reporting, AI feature traffic is counted inside the Web search type of the Search Console Performance report, not shown as a separate line 1. So expect to infer the effect from your prompt log and from referral data rather than a dedicated report.

Skip this work on pages with no original fact to offer, pages that are not indexed, and pages whose job is a transaction, such as a checkout. Fix access first. No one can guarantee a citation, and an edit that helps one engine may do nothing in another.

Frequently asked questions

What kind of content do AI search engines cite most?

The strongest evidence we found is about passages, not formats: AI Overviews often cite pages that match one sub-question, even when the page ranks outside the top 10 4. Ahrefs suggests checking which formats (definitions, stats, step-by-step guides, video) show up in the answers for your topic, but its study does not show one format is cited more.

Do I need schema markup or llms.txt to be quoted?

For Google's AI features, Google says no special markup, AI text files or extra technical steps are needed beyond being indexed and eligible for a snippet 1. Other assistants have not published the same promise. Markup can still help clarity, but good writing and access come first.

Does adding statistics and quotes really increase citations?

In one academic test, adding quotations and statistics raised a visibility score by roughly 41% and 31% over baseline 3. The test used a simplified engine, so treat it as a strong hint. Only add statistics you can source, because fake ones harm trust.

Does updating a page make AI engines quote it more?

Cited URLs were fresher on average than organic results in Ahrefs' study of 16.975 million URLs, yet they were still about 2.9 years old 5. Update pages when facts change and show the real date. Do not change a date without changing the content.

How long before I see results?

No reliable public benchmark exists for this, and results vary by engine and site. Re-test your prompt list at two and four weeks, with three runs each. If nothing changes, check crawl access and whether the page is indexed before rewriting again.

About the author

Rahil Jain

Rahil Jain is an AI SEO (answer engine and generative engine optimisation) consultant and co-founder of Chalk Labs.

LinkedIn · Method and credentials

Sources

  1. AI features and your website · Google Search CentralOpened October 20, 2026. Supports: No extra requirements, special optimisation, schema or AI text files needed for AI Overviews or AI Mode; page must be indexed and eligible for a snippet; AI feature traffic is counted under the Web search type in Search Console; query fan-out. Page last updated 2025-12-10.
  2. Creating helpful, reliable, people-first content · Google Search CentralOpened October 20, 2026. Supports: Who, how and why questions; bylines linked to author background; disclosing automation or AI use; false credentials as deception; original information, reporting, research or analysis. Page last updated 2026-10-05.
  3. GEO: Generative Engine Optimization · arXiv (Aggarwal et al.; KDD 2024)Opened October 20, 2026. Supports: GEO-bench of 10,000 queries across 25 domains; simplified engine using top five Google results and gpt-3.5-turbo; scores for quotation addition (27.2), statistics addition (25.2), fluency optimisation (24.7), cite sources (24.6) and keyword stuffing (17.7) against a 19.3 baseline; rank-1 versus rank-5 effect of citing sources.
  4. How often do AI Overview citations come from the top 10? (863K SERP study) · AhrefsOpened October 20, 2026. Supports: 37.9% of cited URLs ranked in the first 10 SERP blocks across 863,000 keyword results and about 4 million AI Overview URLs (published 2 March 2026); fan-out offered as a likely explanation; advice to check which formats appear in answers.
  5. New Study: AI Assistants Prefer to Cite 'Fresher' Content (17 Million Citations Analyzed) · AhrefsOpened October 20, 2026. Supports: 16.975 million cited URLs; about 1,064 days average since first publication for cited URLs versus about 1,430 for organic (the article gives 1,432 in text and 1,416 in one table); ChatGPT strongest preference for newer content; cited content still about 2.9 years old; warning against changing dates without changing content. Published 28 July 2025.
  6. Architecting and evaluating an AI-first search API · PerplexityOpened October 20, 2026. Supports: Perplexity breaks documents into self-contained spans that are individually retrieved and ranked (September 2025 engineering post).
  7. A guide to Google Search ranking systems · Google Search CentralOpened October 20, 2026. Supports: Passage ranking is a current AI-based ranking system that identifies individual sections or passages of a web page. Page last updated 2025-12-10.
How this page was made. Built on Rahil Jain's AI SEO method and drafted with AI research assistance. Every statistic links to the page we read it on (opened October 20, 2026 or nearby). Published October 20, 2026; last updated October 20, 2026. Chalk Labs sells AI SEO services, so we have a commercial interest in this topic. Spotted an error? Tell us and we will fix it.