How to get cited in AI search

Why Being Cited Is Not the Same as Ranking

Classic search sends a person to your page and lets your page do the selling. AI search does not. An assistant reads a handful of sources, writes one answer, and names the sources it leaned on. If you are not one of those named sources, you are invisible even when you rank on page one of Google for the same question.

That is the shift most businesses have not adjusted to. The goal is no longer only "rank for the keyword." It is "be the source the answer is built from." Those two goals overlap, but they are not identical, and the second one has different rules.

1. Answer the Question in the First 60 Words

An assistant reading your page is looking for a passage it can lift. If your answer arrives in paragraph nine, after a scene-setting introduction and a story about your founder, the model may never reach it. Lead with a direct, complete answer, then earn the reader's attention with the depth underneath.

Practically: state the answer as a standalone sentence that makes sense with no surrounding context. If a sentence only works when you have already read the two before it, it will not be quoted.

2. Use Real Numbers, Dates and Named Specifics

Generic advice is not citable because any of a thousand pages could have supplied it. Specifics are citable because they can only have come from you. "Reduce your software costs" is not a citation. "Replacing 15 subscriptions with one platform cut this client's monthly software spend from $2,400 to $310" is.

Include the year, the figure, the tool name, the timeframe. Assistants reward precision because precision is checkable, and a checkable claim is a safer claim for the model to repeat.

3. Structure Headings as the Questions People Actually Ask

Write headings the way a person types a question, not the way a marketing team writes a section label. "How much does custom business software cost?" outperforms "Investment considerations" every time, because it matches the query the assistant is resolving.

Then answer that heading immediately below it, in that section, without deferring to a later part of the page. Each section should stand on its own as a small complete answer.

4. Put Comparisons in Tables

Comparison questions are the highest-volume question type in AI search, and tables are the easiest structure for a model to parse and reproduce accurately. If your page compares options, costs, or approaches, put the comparison in an actual table rather than describing it in prose.

5. Make Your Expertise Verifiable

Models weigh who is speaking. A named author with a real background, a founding date, a company identity, and a track record is treated differently from an anonymous page. That is why author details, business identity markup, and specific claims about experience matter more in AI search than they did in classic SEO.

This is also why thin, unsigned, AI-written filler does poorly here. It has nothing verifiable behind it, so there is no reason for an assistant to trust it over the alternative.

The pages that get cited are the ones that would still be useful if you stripped away the branding. Be the most specific, most checkable answer on the internet for a narrow question.

6. Let the Crawlers In

None of the above matters if the assistants cannot read your site. Two things break this quietly. First, a robots file that blocks AI crawlers such as GPTBot, ClaudeBot, PerplexityBot and OAI-SearchBot. Second, a site that renders its content only in the browser, so a crawler fetching the page receives an empty shell.

Server-rendered pages and an explicitly permissive robots file are the baseline. Many modern site builders fail the second test without ever telling the owner, which is why a site can look perfect to a visitor and be blank to an assistant.

7. Cover the Whole Question, Not Just the Keyword

A single question rarely arrives alone. Someone asking how to get cited by AI will also ask what tools to use, how long it takes, and how to measure it. Pages that address the adjacent questions in one place get cited across the whole cluster, because the assistant keeps finding its next answer on the same source.

How to Tell Whether It Is Working

Ask the assistants directly and repeatedly. Query ChatGPT, Claude, Gemini and Perplexity with the questions you want to own, and record whether you are named and what you are named for. Track referral traffic from assistant domains separately from Google, because it converts differently: fewer visits, far higher intent, since the visitor arrives already recommended.

Expect a slower feedback loop than classic rankings. Models refresh their view of your site on their own schedule, so a change made today may surface in answers weeks later.