Search is splitting into two behaviours. People still click through to websites. They also, increasingly, read an answer assembled from several websites and never click anything.
The second behaviour is not going away, and it changes what winning looks like. Ranking first is no longer sufficient, because the answer sits above you and may not mention you. The goal becomes being one of the sources the answer is built from.
The good news is that this is not a separate discipline requiring separate infrastructure. It is mostly a change in how pages are written, plus some technical work you should have done anyway.
How These Systems Choose Sources
Different products, broadly the same mechanism.
Google AI Overviews generate an answer from pages already ranking well for the query and related queries, then cite a handful.
ChatGPT search and Perplexity run live retrieval, pull several pages, and synthesise, citing as they go.
Claude, Gemini, and Copilot do a version of the same thing when connected to search.
Three consequences follow.
Conventional ranking still matters. These systems mostly draw from pages that already rank. An invisible page is not a candidate. SEO is the entry ticket, not the whole game.
They extract passages, not pages. The unit being retrieved is a paragraph or a section that answers something specific. A page can be cited for one section while the rest is ignored.
They favour text that stands alone. A paragraph that only makes sense after reading the three above it is hard to lift cleanly. A paragraph that answers a question completely on its own is easy.
That third point is the single most actionable thing in this article.
Write Passages That Can Be Lifted
The structural change is small and the effect is large: answer the question immediately, then explain.
Most business writing does the opposite. It sets context, builds up, and delivers the answer in the fourth paragraph. That is fine for a human reading top to bottom and useless for a retrieval system scoring passages independently.
Instead of:
Many Indian sellers ask about fulfilment costs. There are several considerations. Amazon's fee structure has evolved considerably over recent years and varies across categories. In this section we will explore the various components...
Write:
Amazon India charges FBA sellers four unavoidable fees: a referral fee of 2 to 17 percent of selling price by category, a fulfilment fee of ₹27 to ₹180 or more by size tier, a monthly storage fee of roughly ₹35 to ₹50 per cubic foot, and a closing fee of ₹5 to ₹60 per unit.
The second version is a complete answer with concrete numbers. It can be cited without any surrounding context, and a system deciding between sources will prefer it every time.
The Practices That Work
Use question-shaped headings
Make H2 and H3 headings the actual questions people ask, then answer directly beneath. "What does Amazon FBA cost in India?" beats "Fee Structure". This maps a page onto real queries and gives retrieval a clean boundary around each answer.
Be specific and quantitative
Numbers, ranges, dates, named tools, and named steps. "FBA fees typically run 15 to 30 percent of selling price" is citable. "FBA fees can be significant" is not, and no system will choose it when an alternative has the number.
Vagueness is the most common reason good content is passed over.
Keep facts current and dated
These systems weight freshness, and stale facts are actively harmful because a wrong number that gets cited is worse than no citation. Put a visible last-updated date on anything with prices, fees, or regulations in it, and actually update them.
Add structured data
FAQPage, HowTo, Article, Organization, and LocalBusiness give machines unambiguous facts instead of inferred ones.
One warning, because this is a real penalty rather than a missed opportunity: FAQ markup must correspond to text visible on the page. Marking up questions that a reader cannot see is a guidelines violation. It is also easy to do by accident when schema is generated separately from content, and it is worth an automated check in your build.
Build entity clarity
These systems reason about entities, not just keywords. They need to establish that your business is a real thing with a consistent identity.
- Consistent name, address, and description everywhere
- Organization schema on the site
- A substantial About page with real people on it
- Named authors with credentials rather than "Admin"
- Consistent details across LinkedIn, directories, and your own site
Let the crawlers in
An AI crawler blocked in robots.txt cannot cite you. Decide deliberately whether to allow GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and the rest.
There is a genuine trade-off. Allowing them means your content trains and informs models that may answer without sending a click. Blocking them guarantees you are never the cited source. For most businesses that treat content as marketing rather than as the product, visibility is worth more than the content is.
Check your robots.txt and confirm it reflects an actual decision rather than a default.
What Does Not Work
Keyword stuffing. Retrieval works on meaning, not term frequency. Repetition looks like low quality.
Thin AI-generated filler. These systems are unusually good at recognising content with no original substance, since it resembles their own output with nothing added.
Hiding the answer to force scrolling. Burying a conclusion under 800 words of preamble means the passage never gets extracted. The engagement metric you protected costs you the citation.
Marking up content you have not written. Schema for invisible FAQs is the recurring version of this, and it is a violation.
Measuring Something That Barely Reports
This is the genuinely unsolved part. Be honest about it rather than buying a tool that claims otherwise.
What you can do:
- Ask the engines directly. Query ChatGPT, Perplexity, and Google for the questions your customers ask. Note who gets cited. Repeat monthly. Crude, manual, and the most reliable signal available.
- Watch for impressions without clicks in Search Console. Rising impressions with flat clicks on informational queries often means an AI Overview is answering above you. Sometimes you are cited in it.
- Watch referral traffic from
chat.openai.comandperplexity.ai. Small numbers today, and they are growing. - Track branded search volume. If people encounter your name in AI answers, some fraction later search for you directly. This is the clearest downstream evidence that citation is happening.
The Uncomfortable Part
Some of this traffic is not coming back. If a query has a short factual answer, the answer will increasingly be given without a click, and no amount of optimisation changes that.
What follows from that is a shift in what content is for. Pages whose only value was answering a quick factual question will lose traffic. Pages offering judgement, experience, original data, tools, and specifics that only somebody who does the work would know will keep it, because those are the pages that get cited and the pages people click through to read properly.
That was always the better content strategy. AI search has made it the only one.
Where to Start
- Take your ten highest-value pages.
- Rewrite the headings as the questions customers actually ask.
- Make the first sentence under each heading a complete, specific answer.
- Add numbers where you currently have adjectives.
- Add a visible last-updated date and keep it true.
- Confirm your robots.txt reflects a real decision on AI crawlers.
That covers most of the available gain. The rest is the same work as a technical SEO audit.
Want to Be the Cited Source?
Our digital marketing team restructures content for both conventional search and answer engines, including the structured data and entity work underneath.