AI models recommend brands they can check. A recommendation is not a ranking of websites. It is a shortlist a model builds from what it already knows and what it can pull from the live web in the moment, and then it names the brands whose facts fit the question and hold up against the pages it read. If your brand is missing from both of those places, or if what the model finds about you disagrees from one place to the next, it names a competitor it can check instead.
There is no form to submit your brand for recommendation, and no seller can promise you a place in an answer. Placement is not theirs to give. What you can control is whether the model can reach your site, whether it finds the same facts about you everywhere it looks, and whether those facts answer the exact question being asked.
The two paths into an answer
A model knows your brand through two paths, and they reward different work.
The first path is training. A model learns from large collections of text gathered before it was trained, and that learned knowledge shapes what it believes about a brand before anyone asks. Training rewards brands that have been written about widely and consistently over time: your own site, press coverage, review pages, and listings that all describe you the same way. You cannot edit training directly. You can only change what gets written about you, and keep it consistent.
The second path is live retrieval. When someone asks a question, an assistant fetches current pages and ties its answer to them. OpenAI documents this split in its own crawler pages: GPTBot gathers content that may be used in training, while OAI-SearchBot surfaces sites in search results, and each one is controlled separately in robots.txt. Perplexity documents the same split: PerplexityBot surfaces sites in search results and is not used to crawl content for training. Live retrieval rewards pages that are crawlable right now, written in plain HTML, and specific enough to answer the question.
A brand can be strong on one path and invisible on the other. A new brand with no press history may still get cited from live pages. An established brand with a blocked site may still be named from training, but with stale facts. The work is to cover both.
What Google documents about its own answers
Google is the one engine that has published a full description of how its answers are built, in its guide to optimizing for generative AI features on Search. That guide covers AI Overviews and AI Mode, and it says the features are rooted in the core Search ranking and quality systems, not built beside them.
Two techniques do the work. Retrieval-augmented generation, which Google also calls grounding, uses the core ranking systems to retrieve relevant, current pages from the Search index, and the system then reviews the information from those pages to generate a response with prominent links to the pages that support it. Query fan-out is a set of concurrent, related queries the model generates to fetch more results for the question. Google gives the example of a question about fixing a weedy lawn fanning out into searches for herbicides, chemical-free removal, and prevention.
Two practical points follow from that. First, a page has to be indexed and eligible to be shown in Search with a snippet to be eligible for those generative features. A page Google has never indexed cannot ground an answer. Second, fan-out means your page can be pulled in for a subtopic you never targeted, as long as one section answers that subtopic clearly. Google also warns against building a separate page for every possible query variation to manipulate rankings, since that can violate its scaled content abuse spam policy.
The same guide says structured data is not required for generative AI search, and there is no special schema markup you need to add for it. Schema still matters as part of general SEO, and it still helps other engines read your facts, but it is not a ticket into an AI Overview on its own.
The shortlist: what has to be true
When a model builds its shortlist, four things have to line up.
- The model can find you. Your pages are crawlable, indexed, and written so a retrieval system can match them to the question. If a crawler is blocked or the content only appears after JavaScript runs, many engines never see it.
- The model can tell who you are. Your brand name, what you sell, and where you operate are stated the same way on your site, in your structured data, and on the pages that list you. When the name changes spelling from place to place, the model may treat you as several different brands, or as none at all.
- The facts fit the question. A recommendation answers a question with limits: a budget, a use case, a location, a feature. The brands that get named are the ones whose pages state those facts plainly, with a price, a size, a stock status, or a service area the model can quote.
- Someone else agrees with you. A claim that only appears on your own site is weaker than a claim that also appears on independent pages: a review site, a press article, a retailer listing, a roundup. Third-party pages carry weight because they are the corroboration a model uses when it cannot check the fact itself.
No single one of these guarantees a mention. Together they are what makes a brand recommendable: findable, identifiable, specific, and corroborated.
What to fix first
Work in this order, because each step depends on the one before it.
- Confirm the crawlers can reach you. Check your robots.txt and your edge security for the search crawlers that feed live answers, and confirm in your server logs that they are actually arriving. A brand that is blocked at the edge is invisible on the live path no matter how good its content is.
- Make your identity consistent. Pick one spelling of your brand name and one description of what you sell, and use them on your homepage, your About page, your product pages, your listings, and your Organization markup. For the full treatment, see our guide to adding Organization schema for brand clarity.
- Put the facts buyers ask about on the page. Price, availability, sizes, service area, shipping, returns. Write them in the visible page text, in plain HTML, not only inside a script or behind a button. A model can only recommend what it can read.
- Answer the question, not the keyword. Write one page per real question a buyer asks, open with a direct answer, and keep each section self-contained so it can be lifted on its own. Fan-out means a section that answers a subtopic well can be cited for that subtopic.
- Earn independent mentions. You cannot manufacture these, and Google says seeking inauthentic mentions is not helpful. What you can do is give press, reviewers, and roundup writers something specific to quote: a first-hand test, a clear comparison, a fact no one else publishes.
Say you run a trail shoe brand
Say you run a small brand that sells trail running shoes. A shopper asks an assistant for a waterproof trail shoe under a set budget for wet winter runs. The model fans that question out into subtopics: waterproof trail shoes, winter grip, and shoes in that price range. Your product page states the price, the waterproof lining, and the lug depth in plain text, and a running magazine roundup lists your shoe with the same price. A competitor has a prettier page, but its price only appears in an image and its site blocks the search crawler. The model can check your facts and cannot check theirs, so it names you and not them. Nothing about that outcome required a submission form. It required reachable pages, consistent facts, and one independent page that agreed.
What this guide does not cover
This guide covers how a recommendation gets made. It does not cover the per-engine playbooks for getting each assistant to find you, the full technical audit that checks every crawler and every page, or the transaction layer that lets an agent complete a purchase once your brand is recommended. Those are separate jobs, and this guide is the map, not the whole trip.
Check your own brand now
Run the free scan. It takes seconds, no signup, and shows whether AI crawlers can read your brand today: aizii.ai/aeo-scan
Run the free scanSources
- Google Search Central, Optimizing your website for generative AI features on Google Search (specific): https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
- OpenAI, Overview of OpenAI crawlers (specific): https://platform.openai.com/docs/bots
- Perplexity, Perplexity crawlers (specific): https://docs.perplexity.ai/guides/bots
- Google Search Central, Organization structured data (specific): https://developers.google.com/search/docs/appearance/structured-data/organization
- Schema.org, Organization type (general standard): https://schema.org/Organization
Note: links marked specific point to the page that supports the claim in this guide. Links marked general docs point to the vendor's general docs section for that topic, not to a page that proves every step. Check the vendor docs for the current path before relying on a menu name or click path in this guide.
