Guides / Get Found

By DLVX · 12 min read · Updated 20 October 2026 · Part 5 of 10 in Get Found

Write pages that AI answers can cite

  • Pages get cited when they answer clearly, carry a fact nobody else has, and can already be found in search.
  • Ranking helps, but most studies find a large share of citations go to pages outside the top 10.
  • Skip the tricks: chunking, FAQ schema, llms.txt and fake update dates do not earn citations.
On this page

What actually gets cited

Start with the plain facts. AI answers on Google pull from pages Google can already index and show as a snippet. Google says it plainly: a page must be indexed and eligible to show with a snippet, there are no additional technical requirements, and inclusion is never guaranteed. So the first rule of being cited is boring. Be a good, findable page.

Do we have to rank on page one to be cited?

It helps a lot, and it is not required. The big studies disagree on how much, mostly because they measure differently. Read them together and the shape is clear.

StudyFindingCaveat
Ahrefs, March 202637.9% of AI Overview citations ranked in the top 10 for the query, 31% ranked outside the top 100.Their parsing changed since a 2025 version that said 76%, so the two are not like for like.
Conductor, September 2026About 168 million citations: a page at position 1 was cited about 24.9% of the time, at position 10 about 10.2%. 57% of citations came from outside the top 10.Vendor data, same searches, correlation.
BrightEdge, September 2025Overlap between AI Overview citations and organic rankings rose from 32.3% to 54.5% in 16 months.Overlap with rankings in general, not the top 10.
Ahrefs, August 2025About 15,000 long-tail queries across several AI engines: on average about 12% of cited URLs were in Google's top 10 for the original query.Long-tail prompts, many engines. Ahrefs suggests fan-out as one explanation; it is not measured.
Grow and Convert, March 2026100 buying prompts in ChatGPT: roughly 40% of cited sources appeared in the first ten pages of Google or Bing for the hidden fan-out queries.One agency, small sample.

Our read: ranking well raises the odds of being cited, but a large share of citations go to pages that rank for something else. Query fan-out is the most likely reason. The engine searches several related questions and pulls a source from each, which is why writing for the searches AI runs matters as much as writing for the one the buyer typed.

Answer first, then earn the rest

A page that gets cited usually answers the question in the first two sentences under the heading, in words a stranger could lift out and repeat. Then it backs the answer up. We write every section so it would still make sense if it were the only part an engine read.

There is a lot of nonsense around this, so be precise about what Google says. Its guide for generative AI features, last updated 10 July 2026, says there is no requirement to break content into tiny pieces and no ideal page length. Chunking your page into 40-word fragments for the robots is not a thing. Clear sections are good writing, and that is the whole reason we use them.

ElementWhat we doWhy
HeadingThe buyer's question, in the buyer's words.It matches how people and engines phrase the search.
First two sentencesThe direct answer, with the key number or condition.That is the part most likely to be lifted.
Next paragraphThe reason, the exception, the "it depends on".This is where trust is earned.
EvidenceA figure, a named source, a dated example, a table.Specifics separate you from the twenty pages that say the same thing.
Next stepA link to the service page or the next question.Visibility only matters if it leads somewhere.

How long should the page be?

As long as the question needs. Some of the most cited pages we read are short reference pages with one table. Some are long guides. Length was not what separated them; usefulness was.

Give engines something nobody else has

If your page says what every other page says, the engine has no reason to pick yours. The single most reliable way to become the source is to own a fact.

  • Original data. A count, a survey, a price index, a before and after. Even a small honest dataset is rarer than you think.
  • Named sources. Link the primary document, not a blog that summarised it, and give its date.
  • Quotes with names. A real person with a real role saying something specific.
  • Worked examples. Numbers that walk through a real case, with the assumptions shown.

The academic support for this comes from the GEO paper by researchers at Princeton, Georgia Tech, IIT Delhi and the Allen Institute, first published in November 2023. They reported that adding statistics, quotations and cited sources could raise visibility by up to 40 percent. Keep it in proportion: they tested a research engine in 2023, not today's ChatGPT or AI Mode, and we have found no independent replication. We treat it as direction, not a forecast.

We practise this ourselves. Our Bangkok study asked Google AI Mode 80 buyer questions, three times each, and published every count and the full method. That page exists because nobody else had the numbers.

Do not invent the evidence

A made-up statistic, an anonymous "study", a quote nobody said: these are the fastest way to lose trust with a buyer and with Google. Google's people-first guidance calls made-up creators and false credentials deceptive. If you do not have the number, write the sentence without it.

Freshness: update facts, never fake dates

Ahrefs analysed about 17 million citations (July 2025). AI-cited URLs were on average 1,064 days old against 1,432 days for organic results, about 25.7 percent younger, and about 13.1 percent younger by last update. ChatGPT leaned fresher; Google AI Overviews did not.

Two things in that study get lost in the retelling. It is a correlation, not a test. And the average cited page was still close to three years old. Old pages get cited all the time when they are still right.

PracticeVerdictWhy
Update the page when a price, rule or fact changesWorksCorrect pages earn trust, and the change gives a real reason for a new modified date.
Show a visible "updated" date that matches the changeWorksReaders check dates. So do engines.
Change the date without changing the contentHypeNo test supports it. It also teaches readers to ignore your dates.
Rewrite old posts every month "for freshness"Skip itEffort with no evidence behind it, and a risk of breaking pages that work.

We run a monthly freshness pass on our clients' most visited guides: old years in the text, prices that changed, dead links, dates that no longer match. That is maintenance, and it is real.

Can AI write these pages?

Can we just have AI write a hundred of these?

You can have AI help write them. You cannot have AI mass-produce them and expect it to hold. Google's guidance on generative AI content, last updated 1 October 2026, says AI can help with research and structure, but generating many pages without adding value may violate the scaled content abuse policy. It also asks you to fact-check everything, including titles, meta descriptions, structured data and image alt text, because models can produce confident errors.

Changed in 2026: that update aligned the guidance with the Search Quality Rater Guidelines' sections on scaled content and low-effort, low-originality content. Google is careful to add that rater guidelines are not a ranking manual. The direction is still unmistakable.

EvidenceWhat it shows
Ahrefs, July 2025, 600,000 ranking pagesThe share of AI-written text had almost no relation to ranking position (correlation 0.011). Most ranking pages were mixed human and AI.
Originality.ai, reported in March 2024A vendor scan of sites hit after Google's March 2024 update: 151 of 175 deindexed sites likely published AI content. The vendor sells AI detection and inferred the penalties.
Glenn Gabe, August 2026 spam updateFour case studies with programmatic and AI-generated sections: sites that lost from about 14,000 to over 200,000 queries (one case is reported as indexed URLs instead). Hand-picked cases, no control group.
Google spam policies, updated 28 August 2026Scaled content abuse covers many pages made mainly to manipulate rankings, no matter how they were created.

Put together: the trouble starts with volume and empty pages, whatever wrote them. We use AI to draft, research and check. A person who knows the subject decides what goes live, adds the facts only the business has, and signs it off. Google's people-first guidance asks you to make clear who made the content, how, and why. We think that is good practice whatever any engine wants.

The page template we use

This is the skeleton we hand to writers for any page meant to be cited. It is not a formula for word counts. It is a list of jobs the page has to do.

Title: the buyer's phrase, under 60 characters, brand last
H1: the human version of the question
First two sentences: the answer, with the key number or condition

H2 (a sub-question buyers ask)
  Answer in one or two sentences
  Why, exceptions, a table or a worked example
  One named, dated source where it matters

H2 (the next sub-question)
  ...

What this costs or how long it takes (if you can say it)
Who this is not for
Updated date that matches the last real change
Link to the service page or the next guide
  1. Collect the questions. From enquiries, sales calls, long Search Console queries and a fan-out harvest.
  2. Pick one intent. One page per buying intent. Variants of the phrasing go into the same page as headings, never into new pages.
  3. Write the answers first. Two sentences each, before any introduction.
  4. Add the evidence only you have. A number, a case, a photo, a price basis.
  5. Have a person check every fact. Especially anything an AI drafted.
  6. Link it in. Two to five links from related pages with descriptive anchors. See internal linking.

Which pages to write first

Nobody has the budget to write for every question. We rank the candidates with three tests, in this order.

  1. Is there buying intent? A question asked by someone close to choosing a supplier beats a question asked out of curiosity. "How much does X cost in Bangkok" beats "what is X".
  2. Do we have something the other answers lack? A real price basis, a dataset, a process we can show, a photo of the actual work. If the answer is "no", the page will be one of twenty, and engines have nineteen alternatives.
  3. Does an existing page already half answer it? Then improve that page. A heading, a better first sentence and a table on a page Google already trusts usually beats a new URL that starts from zero.

What about the pages we already have that nobody reads?

Look before you delete. In our experience the usual problem is not quality but reach: the page is not linked from anywhere a crawler visits often, or its title does not use the words buyers use. Fix those two first. If a page is genuinely useless, merge it into a better one and redirect, so whatever little it earned is kept.

One more filter we apply: no new informational page ships if it does not link to a service or contact page above the fold and once in the body. A cited page that leads nowhere is a gift to whoever the engine recommends next.

What is hype in "content for AI"

TacticVerdictWhat the evidence says
Chunk pages into tiny blocks for AIHypeGoogle: no requirement to break content into tiny pieces (July 2026).
FAQ schema to win AI answersSkip itFAQ rich results stopped showing on 7 May 2026. A visible Q and A for readers is still fine.
Add schema to get citedHypeAhrefs' May 2026 matched test: no measurable lift in AI Mode or ChatGPT citations for pages that added it. See structured data for what schema is good for.
An llms.txt fileSkip itGoogle Search does not use it; SE Ranking found no link to citations across nearly 300,000 domains.
Separate pages for every phrasingAgainst the rulesGoogle names this as scaled content abuse.
A minimum word countSkip itGoogle has no preferred length.
Original data, clear answers, named sourcesWorksThe common thread in every serious study.

Notice the pattern. The tricks that promise a shortcut are the ones the engines either ignore or punish. The work that holds is the same work that makes a page useful to a person deciding whether to buy.

A worked rewrite, then the checklist

A typical service page opens with a paragraph about passion and experience, and gets to the service in the third scroll. This is the same section rewritten the way we would.

BeforeAfter
"We are a passionate team with years of experience delivering world class solutions to businesses of every size."How long does a website rebuild take? Most rebuilds take [your real range, say in weeks] from signed scope to launch. The range depends on how many pages move and whether the content is ready on day one.
No price information."What drives the price: number of templates, languages, integrations and who writes the copy." Plus a link to the pricing page.
Stock photo.A real screenshot of a delivered page, or a table of what is included.

The "after" is not cleverer. It is just answerable. An engine can lift the first two sentences, and a buyer gets what they came for. Put your own real range where the brackets are. If you do not know it, find out before you publish the page.

Checklist

  • The page is indexed and can show a snippet.
  • Each H2 is a real buyer question and is answered in its first two sentences.
  • At least one fact on the page is yours alone.
  • Every external claim links a dated primary source.
  • Dates on the page match the last real change.
  • A person checked every fact, including anything AI drafted.
  • One page per intent, no near-duplicate variants.
  • It links to the next step and gets links from related pages.

Next in the track: Query fan-out: write for the searches AI runs. Related: measure whether it worked, and the start of the track, what works in search and AI visibility in 2026.

Sources and dates

Every external fact in this guide was read on 20 October 2026. Each source shows its publisher and the date it was published or last updated, and we flag what changed in 2026. Studies are their authors' measurements, usually correlations, not promises of results. Our own numbers come from our own Search Console and tools and say so where they appear.

Want this run on your own site? Start with the AI visibility diagnostic, or see the whole Get Found track.