Guide
How a business becomes visible in AI search answers
By Mavrin Labs. Published . Last updated . How this guide was researched and checked.
You cannot make an AI assistant recommend your business, and no firm can. What you can control is whether your pages can be crawled, whether they answer the question plainly, and whether your details agree everywhere. This guide separates the two.
Why a business goes missing from AI answers
Buyers now ask an assistant the question they used to type into a search box, and the answer arrives as a short recommendation rather than a list of links. When your business is not in that answer, the cause is usually ordinary. The site blocks the crawler. The page buries its answer under a scroll of introduction. The business details differ between the site and the listings. The markup describes a page that does not exist.
None of those problems is exotic, and all of them are fixable. The uncomfortable part is what remains after you fix them: the decision about which sources an answer cites belongs to systems nobody outside those companies can see or influence.
There is no route into an AI answer to buy, and no firm that can place you in one. The work removes the reasons you would be skipped.
What is AI search visibility?
AI search visibility is the chance that a system writing an answer finds your pages, reads them, and uses them as a source. It is a chance rather than a position, because there is no ranking to occupy and no placement to buy.
Classic search remains the foundation. The engines that write answers crawl the web with the same kind of crawler, read the same pages, and rely on the same signals of what a page is about. A site that ordinary search cannot read is invisible to the answer engines too. A site with clear, answerable pages is a candidate for both.
The practical distinction is granularity. Ordinary search sends a reader to a page. An answer engine takes a passage. That changes what you optimise: the paragraph directly under a heading has to stand alone and be true on its own, because it may travel without the rest of the page around it.
What you can control, and what nobody controls
You control five things: whether crawlers may fetch your pages, whether those pages are indexable, whether each one answers a real question early and plainly, whether your structured data matches the visible content, and whether your business details are the same wherever they appear.
Nobody controls which sources an answer cites, how often an engine re-reads a site, whether a given model has seen your pages at all, or what a system says about your business when it guesses. There is no submission route, no verified placement and no supplier who can arrange one. Mavrin Labs sells no advertising, and no advertising product places a business inside an answer either.
- Yours to control
- Crawl access, indexable pages, an answer in the first paragraph, markup that matches the page, and one consistent set of business details.
- Nobody's to control
- Which sources an answer cites, when an engine re-reads your site, whether a model has seen your pages, and what it says when it guesses.
That is worth stating plainly before any work starts. Technical search work makes a business findable and readable. Whether an answer engine then recommends it depends on systems Mavrin Labs does not control, and any firm promising otherwise is selling something it cannot deliver.
How to prepare a site, in order
The order below matters, because each step is wasted if the one above it is broken.
A plain index helps too. Publish a sitemap for machines, and if you like, a plain-text index of the pages that matter.
Classic search and AI answers compared
| What it depends on | Classic search | AI answers |
|---|---|---|
| Crawl access | required | required |
| Indexable page | required | helps, and nothing replaces it |
| Clear subject per page | strongly rewarded | strongly rewarded |
| Answer-first passages | helps the snippet | central, because a passage is quoted |
| Structured data matching the page | supports rich results | supports machine understanding |
| Consistent business details | supports local results | heavily relied on |
| A route to buy placement | none in organic results | none at all |
| Reported position | available | none exists |
How to measure it without inventing numbers
Measure with a method you repeat, and record the method alongside the result. There is no report that tells you your visibility in AI answers, so anyone offering a score is estimating.
Keep a fixed list of the questions a buyer would ask. Run the list on a fixed date each month, record the answers you receive and note whether your business appeared and in what terms. Keep the raw answers, because the wording tells you which of your pages did the work. Alongside that, watch the search console reports for your site and the referrers in your own analytics, which show when an assistant sends a visitor.
Expect noise. The same question asked twice can return different answers, so treat a single result as an anecdote and the trend over several months as the signal.
What not to do
Hidden text, pages written for crawlers rather than readers, instructions buried in a page that try to tell an AI system what to say, invented reviews and manufactured citations are all manipulation, and the published guidance for site owners treats them that way.
- Do not hide text, whether behind a colour, a zero height or an off-screen position.
- Do not write pages for crawlers that a reader would find worthless.
- Do not plant instructions in a page aimed at the systems reading it.
- Do not create reviews, ratings or citations that nobody gave you.
- Do not publish a page whose markup describes something the page does not show.
Beyond the risk of a penalty, these tactics fail on their own terms. A page that reads as machine-written persuades nobody who arrives on it, and a citation you engineered is worth nothing when the reader clicks through and finds no substance. The durable version of this work is slower and less exciting: fewer pages, each genuinely answering something, kept current and honestly marked up.
What shapes the investment
Three things shape what this work takes. The first is the state of the site: a site that already renders its content in plain markup needs editing, while a site whose pages assemble themselves in the browser may need rebuilding before any of this applies. The second is the number of pages that genuinely deserve to exist, which is usually smaller than a site already has. The third is whether your business details need reconciling across listings, which is tedious rather than difficult.
The investment is agreed in writing after the first review, once those three are known. No figure can be quoted before them without guessing.
Standards behind this guide
Three public references carry the technical part of this guide, all of them opened and read on Thursday, October 8, 2026. The documentation for site owners published by Google explains what makes content helpful and how its systems read pages (helpful content guidance), and it sets out the rules for structured data, including the requirement that markup match visible content (structured data guidelines).
The vocabulary itself is published and maintained in the open, which is what makes markup portable between engines (the vocabulary). The llms.txt convention is a proposal rather than a standard, and it is worth reading as one (the proposal).
Where those sources say nothing, this guide says nothing. There is no published account of how any answer engine weighs its sources, so none is offered here.
Getting started
Begin with the three checks that reveal most: fetch your own robots file and read it, view the raw source of your main page and confirm the text is in it, and ask two assistants a question a buyer would ask. Those three take an afternoon and usually explain the problem.
From there, technical SEO and AI search visibility is the service that covers this work, and the AI search visibility programme sets out how it runs. If your wider question is whether to build, buy, connect or automate, the systems decision framework addresses that instead.
Frequently asked questions
Can anyone get a business into an AI answer?
Nobody can place a business in an AI answer. The systems that write those answers choose their own sources, and no firm has a route in. What you can do is remove the reasons you would be skipped: let the crawlers in, publish pages that answer the question plainly, keep your business details consistent, and make sure your markup matches what a reader sees. That work also helps ordinary search, which is why it is worth doing whatever the answer engines decide.
Do I need a separate strategy for AI search?
A separate strategy is not needed, because the foundations are the same ones classic search rewards. Crawlable pages, a clear subject per page, honest headings, structured data that matches the visible content and consistent business details all serve both. What changes is the emphasis: answer engines quote short passages, so the first paragraph under a heading matters more, and they rely on your details agreeing across the web, so inconsistency hurts more than it used to.
Should I block AI crawlers or allow them?
Blocking or allowing AI crawlers is a business decision, and it has a direct consequence. A crawler you block cannot read your pages, so it cannot cite them. Some owners accept that in exchange for keeping their content out of training data. Others separate the two: allow the crawlers that fetch pages to answer a question now, and refuse the ones that collect text for training. Both positions are legitimate, and the robots file is where you record whichever one you choose.
What is an llms.txt file and do I need one?
An llms.txt file is a proposed plain-text file at the root of a site that lists its main pages for software that reads them. The proposal is not a standard and no engine has committed to using it. Publishing one takes very little effort and risks nothing, and it doubles as a tidy index of what your site actually covers, so it is reasonable to add. It is not a substitute for pages that can be crawled and read.
How do I measure whether AI visibility is improving?
Measure it with a repeated question set and your own referral data. Write down the questions a buyer would ask, run the same list on a set day each month, and record what came back and whether you appeared. Then watch which referrers send visits to your site and which pages they land on. Neither measure is complete, and both move for reasons outside your site, so record what you did as well as what you saw.
Does adding structured data make a page more likely to be cited?
Structured data does not buy a citation. Its job is to describe what is already on the page in a form software can read without guessing: what the page is about, who published it, when it was updated. Markup that claims something the page does not show is a misrepresentation, and the published guidelines treat it as one. Add it to describe the page accurately, and expect it to help machines understand you rather than to promote you.
Will writing more pages help?
Writing more pages helps only when each one answers a question somebody actually asks and says something your other pages do not. Thin pages built around keyword variations compete with each other, dilute the subject of the site and give an answer engine several weak candidates instead of one good one. A smaller set of pages, each genuinely about one thing and kept current, is the version of this work that holds up.
How long before any of this shows a difference?
Nothing here shows a difference on a schedule anyone can promise. The systems involved re-crawl on their own timing, change their behaviour without notice, and give no account of why a source was chosen. What you can control is whether the work is done: the crawl access, the page structure, the markup, the consistency of your details. Record when each change went live, so that when something does move you can say what was in place beforehand.
Who writes these guides
Mavrin Labs delivers client work as a team in mixed roles, led by our founder. The people who write these guides are the people who build the systems described in them, and every page is reviewed before it is published.
Which hours would you take back first?
Send Mavrin Labs one workflow that is costing you the most time, and the reply comes back by email from the people who would build the fix.
Start a conversation