- AI assistants cite pages they find when someone asks, so crawler access and clear answers come first.
- Google says AI Overviews and AI Mode have no extra requirements beyond being indexed and eligible to show with a snippet.
- Measure with a fixed prompt set, and track mentions and citations separately.
- A citation gap analysis shows which of your pages to create or update, and which third-party sources you need to appear on.
- No one can guarantee a citation. Track trends across the prompt set, not single answers.
ChatGPT, Perplexity and Google's AI Overviews build answers from sources they find when someone asks a question, and link to some of them. To get your brand cited, you need pages that answer your buyers' questions clearly enough to quote, crawlers that are allowed to reach them, and mentions on the third-party sites these assistants already rely on for your category. Nobody can guarantee a citation, but you can measure where you stand and close the gaps that are within your control. This is the process I use.
How AI assistants choose what to cite
A language model knows about your brand in two ways. The first is its training data: what it learned from the web before its cut-off date. The second is live retrieval: the assistant searches the web to answer a question, reads a set of pages and cites some of them. Citations come from the second, and that's the part you can influence most directly.
The main surfaces work in similar ways:
- ChatGPT searches the web for many questions and shows the sources it used alongside the answer.
- Perplexity is built around search and lists its sources with every answer.
- Google AI Overviews and AI Mode draw on Google's own index. Google says a page must be indexed and eligible to show with a snippet to appear as a supporting link, and that there are no additional requirements beyond normal SEO best practices.
Google also describes a "query fan-out" technique, where the system runs several related searches across subtopics to build one answer. A buyer's single question becomes several searches, and the answer is assembled from pages that each cover one piece well. That favours pages that answer a specific question directly over broad pages that touch everything lightly.
Step 1: Make sure the assistants can reach you
Start by checking that the crawlers each assistant relies on aren't blocked. Each company documents its own:
- Google: AI Overviews and AI Mode are part of Search, so they rely on Googlebot. Google-Extended is a separate control for training and grounding in Gemini and some of Google's other AI products, and Google states that Google-Extended doesn't affect inclusion in Google Search.
- OpenAI: OAI-SearchBot is the crawler behind ChatGPT's search features. GPTBot is a separate crawler for model training. You can allow one and block the other.
- Perplexity: PerplexityBot is the crawler that surfaces and links sites in Perplexity's answers.
Then check for the things that block crawlers without anyone deciding to:
- robots.txt rules added years ago, or by a plugin
- CDN or firewall settings that challenge or block automated traffic, including AI crawlers
- Key content that only appears after JavaScript runs. Not every AI crawler renders JavaScript the way Googlebot does, so keep the important text in the HTML.
- Snippet controls. Google's documentation points to nosnippet, max-snippet and noindex as the ways to limit what it shows from your pages, and that includes its AI features.
If you're not sure what's blocked or why, a technical SEO audit will find it.
Step 2: Build a prompt set and measure where you stand
AI answers vary by prompt, by user and over time. The way around that is a consistent set of prompts, run the same way each time.
- List your buyers' questions. Use sales calls, support tickets and search data. Cover every stage, from "how do I fix this problem" to "which vendor suits a company like mine" to "Vendor A vs Vendor B".
- Write them the way people ask. Full sentences with context, like "What accounting software works best for a construction company with 40 employees?", not two-word keywords.
- Run them across the assistants. At a minimum, ChatGPT, Perplexity, Gemini and Google AI Overviews.
- Record everything. Whether you're mentioned, whether you're cited with a link, which competitors appear, and the exact URLs cited.
- Repeat on a schedule. Same prompts, same method, so you can see what's changing.
Keep mentions and citations apart. A mention means the answer names you. A citation means it links to a source, which may or may not be your site. Plenty of brands are mentioned in answers that cite someone else's review of them.
Step 3: Find the citation gaps
This is the step most brands skip, and it's where the plan comes from. I classify every cited URL by the type of source, such as a vendor page, review site, comparison article, forum thread or documentation, and by the topic it covers. Then I rank the sources by how much weight they carry across the whole prompt set, and compare where each competitor is cited with where you are.
The gaps fall into three types:
- Owned-page gaps. A competitor's own page gets cited for a topic where you have no page, or a weaker one. The fix is to create or update a page.
- Third-party gaps. A review site, list or comparison article gets cited, and it names competitors but not you. The fix is to get onto that source.
- Topic gaps. A whole theme where competitors show up across many answers and you barely appear. The fix is a group of related pages, not one.
Each gap becomes an item on an update-or-create list, ranked by how much that source shapes the answers.
Step 4: Make your pages easy to cite
Being retrieved isn't the same as being quoted. These are the changes I make to pages that should be earning citations:
- Answer first. Put the direct answer in the first paragraph under each heading, then add the detail.
- One question per section. Use headings that say what the section answers, so a system looking for one subtopic can find it.
- State facts plainly. Who the product is for, how pricing works, what it integrates with, and where it doesn't fit. Vague positioning copy gives an assistant nothing to quote.
- Use lists and tables for comparisons. A structured comparison is easier to extract than one buried in paragraphs.
- Add something only you can say. Your own process, examples, data or product detail. A page that repeats what every other page says gives an assistant little reason to choose it.
- Keep pages current. Update pages when the facts change, and show when they were last revised.
- Make the facts consistent. Your site, review profiles and directory listings should describe you the same way. Schema markup with consistent entity IDs helps machines connect them, though Google says there's no special schema you need to add for its AI features.
Here's the difference in practice, for a made-up accounting software company:
Before: "We help growing businesses take control of their finances with smarter tools."
After: "[Product] is accounting software for Canadian construction companies with 10 to 200 employees. It handles job costing, progress billing and holdbacks, and connects to the payroll tools most contractors already use."
The second version answers "who is it for" and "what does it do" in two sentences an assistant can lift as they are.
Step 5: Earn mentions where the answers come from
AI answers about a software category often lean on third-party sources. Your citation gap analysis tells you which ones matter for your category. Typical targets:
- Review platforms. Keep your profiles complete and current, and ask customers to leave reviews there.
- Comparison and "best of" articles. Contact the publishers of the ones that get cited, and offer accurate, current information about your product.
- Forums and communities. Take part honestly, as yourself, where your buyers ask questions. Planted posts get noticed and do more harm than good.
- Integration and partner directories. Every listing is one more place that describes what you do.
- Your own documentation. Public, indexable help docs get cited for how-to questions, and they show buyers what the product is really like.
Step 6: Re-run and track
Run the same prompt set on a schedule and compare. I track how often you're mentioned, how often you're cited, which of your URLs get cited, and how that compares with each competitor. Check your analytics as well. Some assistants send referral visits you can see, and a "How did you hear about us?" field on your forms catches some of the rest.
Expect uneven movement. Answers vary from one run to the next, so look for trends across the prompt set rather than reacting to a single answer.
What to avoid
- Anyone who guarantees you'll be cited. No one controls what an assistant cites.
- Pages written only for AI crawlers, or text hidden from readers. Google's spam policies cover hidden text and keyword stuffing, and Google's guidance for its AI features says the same Search policies apply there.
- Blocking AI crawlers by accident while you're trying to win citations.
- Chasing every prompt at once. Start with the questions closest to a buying decision.
Where to start
Pick ten questions your buyers ask before they buy, run them in ChatGPT and Perplexity, and write down who gets cited. If competitors keep appearing and you don't, you've found your first gaps. My AI search optimization work starts with a fixed-scope AI visibility audit that does this properly across the major assistants. Get in touch and I'll tell you what I'd look at first.