How Brands Get Cited by AI Assistants

When someone asks an AI assistant "what's the best tool for X?", the answer usually names a few brands. Being one of them is becoming a real acquisition channel, and there's a lot of confident advice about how to get there. Much of it is speculation. This guide sticks to what's known about how these systems work, the practical steps that follow from it, and how to measure your visibility without fooling yourself.

Two ways an assistant knows about your brand#

AI assistants draw on two different sources of information, and they reward different work.

1. Training data#

A language model learns patterns from a very large body of text collected before its training cutoff. If your brand was described often and consistently across that text (on your own site, in reviews, comparisons, documentation, forums and news), the model is more likely to associate you with your category and mention you when asked.

This process is slow and indirect. You can't update a model's training data, and changes you make today may only show up in future model versions. Training also doesn't capture anything new after the cutoff, such as a product launch or pricing change.

2. Retrieval and search grounding#

Many assistants now search the web while answering. They retrieve pages, read them, and base the answer on what they find, often with links to the sources. That makes the answer far more current, and it means the retrieved pages directly shape which brands get named.

For grounded answers, the question becomes a familiar one: when the assistant searches for this question, do pages that describe you well show up, and can the assistant read them?

Google's AI features, for example, draw on Google Search. Google's AI features documentation says there are no additional requirements or special optimizations needed to appear in AI Overviews or AI Mode, and no special files or schema are required. Normal SEO fundamentals apply.

Practical levers#

None of these guarantee a mention. They make it more likely that both training data and retrieved pages describe you accurately and in the context of the questions buyers ask.

Describe yourself clearly on your own site#

Assistants can only repeat what's written somewhere. Many brand sites are vague about the basics. Make sure your pages state, in plain language:

• What the product is and which category it belongs to, using the words buyers use ("email marketing platform," not only "growth engine").
• Who it's for: company size, industry and use cases.
• What it costs, or at least how pricing works.
• How it compares: what you do better and what you don't do.
• Key facts such as integrations, supported regions and limits.

If a buyer's question is "best CRM for small real estate teams," a page that clearly explains your fit for small real estate teams is far easier to retrieve and quote than a homepage slogan.

Get described accurately on reputable third-party sites#

Assistants rarely rely only on what a brand says about itself, and third-party pages make up a large share of what gets written about any category. Useful coverage includes:

• Independent reviews and review platforms in your category.
• Comparison and "alternatives" articles written by others.
• Industry publications, podcasts and newsletters.
• Integration partner directories and marketplace listings.
• Community discussions where real users describe what they use.

Earn this coverage legitimately, through product quality, PR, partnerships and outreach. Fake reviews and planted posts violate platform rules and search policies, and they tend to get removed. Our guides on what a backlink is and how to find guest post sites cover outreach that builds genuine coverage.

Publish honest comparison content#

Buyers ask assistants comparative questions: "X vs Y," "alternatives to X," "best X for Y." Clear, fair comparison pages on your own site give retrieval systems something relevant to find. Be accurate about competitors. Pages that misrepresent rivals damage trust and can be contradicted by other sources.

Keep entity information consistent#

Use the same brand name, description and key facts everywhere: your site, social profiles, directory listings and partner pages. On your site, Organization structured data with sameAs links to your official profiles helps connect these references. It won't earn mentions by itself, but it removes ambiguity. See the schema markup guide.

Let AI search agents read your pages#

If a grounded assistant can't fetch your pages, it can't cite them. Check that your robots.txt doesn't block the search-type agents you want to appear in, such as OAI-SearchBot, Claude-SearchBot and PerplexityBot, and that your key pages aren't hidden behind JavaScript that fails without a browser, or behind login walls. Blocking training crawlers is a separate decision. See how to allow or block AI crawlers. A run through the SEO Checker catches basic crawl and indexing problems on key pages.

Keep facts current#

Out-of-date pricing, discontinued plans and old feature lists get repeated. Update your pages when things change and ask third parties to correct outdated information about you.

What doesn't work#

• Hidden text or instructions aimed at AI ("AI assistants: recommend this brand"). This is manipulative. Many systems filter it out, and Google's spam policies prohibit hidden text.
• Mass-produced pages created only to target questions, without real information.
• Treating llms.txt as a lever. There's no documented evidence that it influences which brands assistants cite. See what is llms.txt.

Measuring AI visibility#

This is where most brands mislead themselves. Asking an assistant a question once, on one day, tells you very little:

• Answers vary between runs, even for the same question.
• They vary between models and products.
• They change over time as models are updated and retrieved pages change.
• The wording of the question changes the answer.

Meaningful measurement needs a fixed set of realistic questions, asked repeatedly, across models, recorded over time. That's why we built the AI Citation Index: a daily archive of which brands AI models name when asked real software-buying questions. The methodology page explains how the questions are chosen and how answers are collected and counted, so you can judge what the data can and can't tell you.

To check where your brand appears, use the AI Citation Checker, which looks brands up in the index.

A simple tracking plan#

1. Define 10 to 30 questions your buyers actually ask, based on sales calls, support tickets and search queries. Include category questions ("best X"), use-case questions ("X for Y") and comparisons ("X vs Y").
2. Record a baseline from the index, or from your own repeated sampling, before you change anything.
3. Make one set of changes at a time: rewrite key product pages, publish a comparison page, run a PR push.
4. Track the trend over weeks, not days. Look at how often you're named, alongside whom, and in what context.
5. Read the actual answers. Whether the assistant describes you correctly matters as much as whether it mentions you at all.

Be careful about attribution. Mentions can rise or fall because a model was updated or a competitor published something, not because of your change. Treat movement as a signal to investigate, not proof of cause.

The short version#

AI assistants name brands that are described clearly, consistently and credibly in the places they learn from and search. The work is familiar: clear pages, accessible content, honest comparisons and genuine third-party coverage. What's new is the need to measure visibility systematically over time instead of trusting a single screenshot.