What's AI Saying About Your Brand?

More takeaways from Ethan Smith's talk at Ahrefs Evolve

In our last post, we shared what Ethan Smith of Graphite taught us about brand search in AI: your brand story is being written by the crowd, and AI repeats whatever the crowd says most.

Smith had more to say, and it's worth its own post. Back at the office, we filled the whiteboard again. This time the notes answer a more practical question: when someone asks AI about your brand, where does the answer actually come from, and how long does it take to change it?

The three questions buyers ask about your brand

When people research a brand in AI, they tend to ask some version of three questions:

  1. What is [your brand] best known for?
  2. What are [your brand's] best attributes?
  3. Is [your brand] worth it?

Each one maps to a stage of the buying decision. The first is awareness: what do you do? The second is evaluation: what makes you good? The third is the decision itself: should I spend my money here?

Ask all three about your own brand in ChatGPT, Gemini and Perplexity. Then ask them about your top competitors. The answers are your brand's reputation as AI currently tells it, and they're being read by prospects right now.

Only 7% of AI answers cite a source

This was the number that stopped us. According to Smith, only about 7% of AI answers include citations. The other 93% have none.

That matters because most of the AI visibility conversation focuses on citations: getting your page linked as a source in an AI answer. Citations are valuable, but they are the exception, not the rule. For most answers, AI isn't looking anything up. It's answering from memory.

Train that memory with content you seed.

AI prefers its base model

That memory is the base model: everything the AI learned during training, from the huge body of web content it read. When AI can answer a question from its base model, it usually does. Searching the live web is slower and costs more, so it saves that for questions where it needs fresh information.

Brand questions like "what is [brand] best known for?" are exactly the kind AI is comfortable answering from memory. So for most of the people asking about you, the answer comes from what the model already believes about your brand, not from your latest blog post.

The base model updates every 4 to 6 months

Smith estimated that AI takes 4 to 6 months to update its base model. In practice, what AI says about your brand today reflects what the web said about you months ago.

That cuts both ways. If you fix a reputation problem today, AI may keep repeating the old story for a while. But if you build a strong, consistent body of content now, it works its way into the next update and keeps paying off long after it's published. Brand building for AI is a long game, and the brands that start early get ahead of the next training cycle.

The answer: brand-reaffirming content

If AI mostly answers from memory, and that memory only refreshes a few times a year, the goal is simple: make sure the web says the right things about you, consistently, before the next update.

We call this brand-reaffirming content. It's content, on your site and across the web, that answers those three questions clearly and says the same true things about your brand again and again.

  • Answer "best known for" in plain words. Decide the one or two things you want to be known for and state them clearly on your homepage, About page and product pages.
  • Spell out your best attributes. Use specific, quotable facts: what you do, who it's for, what it costs, what makes it different. Vague marketing language gives AI nothing to repeat.
  • Make the case that you're worth it. Customer stories, reviews, case studies and comparisons give AI the evidence it needs to answer the value question in your favor.
  • Repeat it off your site. Your own claims carry more weight when other sites back them up. Guest posts, news coverage, roundups and reviews that describe your brand the same way all reinforce the same story.
  • Stay consistent over time. One campaign won't move a base model. Steady coverage, month after month, is what builds a body of evidence big enough to show up in the next training cycle.

This is where we focus at Page One Power. Our editorial links now go out in content formats AI already relies on, such as answer-first articles, listicles, versus posts, comparison guides and expert profiles, each written by authors who meet E-E-A-T standards. Every placement earns a quality link and reaffirms what your brand is best known for on sites AI trusts.

For reach, our Media Blitz turns a data study on your site into coverage syndicated across hundreds of news outlets, each with a link back to you. A PR Drop follows with brand mentions across 500 to 800 outlets. Together with editorial links, they put the same story about your brand in front of AI in many places at once, and keep it there through the next update.

The bottom line

When someone asks AI about your brand, 93% of the time it answers from what it already believes. That belief was formed months ago, from everything the web said about you.

You can't edit an AI's memory directly. But you can shape what it learns next. Publish brand-reaffirming content, earn mentions that say the same thing, and keep it up. By the next update, AI will be telling your story the way you'd tell it.

Jon Ball

Jon Ball

Jon Ball is the CEO of Page One Power and co-Editor-in-Chief for Linkarati. Jon regularly speaks at SEO conferences and tradeshows all around North America. He is also a frequent contributor to Search Engine Watch, Search Engine Journal, Search Engine Land, Search Engine People and the SEMRush blog. Jon lives with his family in Boise, ID.

Never Miss an Insight

Join thousands of marketers who receive our latest articles, videos, and link building insights each month.