When someone asks an assistant for the best option in your category, it does not consult a ranking. It reads a handful of pages, finds the names that appear across more than one of them, and repeats those.
That single mechanic explains almost everything about getting recommended. You are not optimising for a ranking algorithm. You are trying to be the name that appears on the pages the model reads.
What the assistants actually do
They read a small number of sources. Typically three to eight pages retrieved live for the query, not an index of the whole web.
They favour corroboration. A business named on one page is a mention. A business named on five independent pages is a fact, as far as the model is concerned.
They prefer pages that are easy to extract from. Clear lists, named entities, explicit comparisons and plain sentences beat narrative prose, because the answer has to be assembled in one pass.
They cite recent pages disproportionately. A 2026 roundup outranks a better 2023 one for most commercial queries.
Step 1. Find out where you stand today
Ask the assistants the questions your buyers ask. "Best [category] for [audience]", "alternatives to [competitor]", "[category] tools for [use case]". Do it in ChatGPT, Gemini, Perplexity and Claude, because they retrieve differently and the answers diverge more than people expect.
Write down the sources cited, not just the names recommended. Those cited URLs are your target list for everything that follows. This is the single most useful hour in the whole process.
Step 2. Get onto the pages that are already being cited
Take the URLs from step 1 and work out how to appear on each. Most are listicles maintained by someone who updates them, which means an email works. If you are unsure what a page worth citing looks like, our roundup of blog examples covers the ones that get referenced.
What to send: a note about what is out of date on their page, your entry written in their format so it can be pasted in, and a link to a page that backs up the claim. Not a request to be added because you exist.
This is the highest-leverage action available and almost nobody does it, because it looks like link building rather than AI optimisation. It is neither. It is source placement.
Step 3. Be describable in one sentence
Models repeat descriptions rather than inventing them. If the sentence describing your business differs on every page, no consistent description exists to repeat.
Write the sentence once, put it in your homepage title, your meta description, your about page, your directory listings and your social bios, and keep it identical. Category first, differentiator second, audience third.
Step 4. Publish comparison pages that name competitors
Queries like "X versus Y" and "alternatives to X" are among the most common commercial prompts, and the pages answering them are what gets retrieved.
Be fair to the competitor. A comparison page that finds nothing good in the alternative is discounted by readers and reads as promotional to a model summarising it. Say plainly where the other product wins.
Step 5. Make the facts extractable
Put pricing in a table, not in a paragraph. Put the feature list in a list. Name the integrations explicitly. Answer the obvious question in the first sentence under the heading that asks it. The same formatting wins in classic search, which our guide to how to get featured snippets covers.
Test it yourself: paste your page into an assistant and ask it to state your pricing and your three differentiators. If it gets any of it wrong, the page is not extractable and no amount of traffic fixes that.
Step 6. Get third-party validation with structure
Review platforms, directories and comparison sites are heavily retrieved because they are structured and current. G2, Capterra, Trustpilot and the category-specific directories in your niche.
Volume and recency both matter. Forty reviews from this year read as an active business. Four hundred from 2022 read as a business that peaked.
Step 7. Publish original numbers
Assistants cite statistics constantly and they need a source for each one. A page of original, clearly sourced numbers about your industry is the most reliably cited asset you can build.
Format it for extraction: one claim per line, the method stated, the date visible, and no burying the number inside a paragraph of context.
Step 8. Keep your dates current
A page dated 2024 competing against one dated 2026 loses on most commercial queries regardless of which is better. Update the content and the date together, and never only the date.
Set a review cycle for the pages that matter, quarterly for commercial pages. This is boring and it works.
Step 9. Answer the questions in public
Reddit, Quora, industry forums and community Slacks are retrieved heavily, particularly for "best" and "recommendation" queries where Google itself surfaces forum results.
Participate as yourself, disclose the affiliation, and be useful when your product is not the answer. Anything else is identifiable and it costs more than it earns.
Step 10. Measure it monthly
Re-run the step 1 prompts on a schedule and log which names appear and which sources are cited. Movement here is slow and the only way to see it is to have last month's answers written down.
Track two things: whether you appear, and whether the sources citing you changed. The second predicts the first.
What does not work
Keyword stuffing for AI. Retrieval is semantic. Repeating a phrase does not increase the chance of extraction and it makes the page worse to summarise.
Hidden text and prompt injection in page content. Instructions aimed at a model, embedded in a page, are a reputational risk and are increasingly filtered.
Buying placements on link-farm listicles. Pages that exist to sell placements are not the pages being retrieved, because nobody else cites them either.
Waiting. Six months of the pages in step 1 being maintained by other people, without your name in them, is six months of a compounding disadvantage.
Where this fits with normal SEO
Almost entirely the same work, weighted differently. Being cited by assistants and ranking in search both come down to being present, correct and corroborated across the pages that cover your category.
The genuine difference is that search rewards being the best page and assistants reward being on the most pages. That changes where you spend the effort: less on one perfect asset, more on being present in every source that answers the question.
Distribb's AI visibility side does the mechanical part of step 1 and step 2: it runs the prompts your buyers would ask, records which sources get cited, and produces the list of pages you are missing from with the pitch angle for each.
Honest limitation: it cannot send the email for you and it cannot make a publisher add you. The outreach and the relationship are yours, and if your product is genuinely not competitive in the category, no visibility tool changes what the ranking pages say about it.
See what the AI search visibility platform reports on.
Related reading
For the search half of the same problem, read our guide to semantic search SEO. To build the corroborating asset, see how to publish a statistics page, and for structuring a topic properly, topic clusters.
FAQ
How long does it take to start appearing in AI answers? Weeks rather than months once you appear on the retrieved sources, because there is no index to wait for. Getting onto those sources is the slow part, typically one to three months of outreach.
Does my own website matter, or only third-party pages? Both, differently. Third-party pages get you named. Your own pages decide whether the description is accurate once you are. A business named correctly with wrong pricing has a site problem, not a visibility problem.
Do I need a separate AI SEO strategy? No. You need your existing content strategy weighted toward comparison pages, original data, extractable formatting and third-party presence. The channel is new, the work mostly is not.
Why does ChatGPT recommend a competitor with a worse product? Because they are on more of the pages it read. That is frustrating and it is also actionable, which is more than can be said for most ranking problems.