AI search is a brand problem

Read time: 4 mins 57 secs

Presented by Tracksuit

In Tracksuit's survey of 723 marketing and finance leaders, 96% of senior marketers say they can make the financial case for brand. The majority also say they don't always have the tools to back it up.

Always-on tracking closes that gap, with brand health numbers ready the day finance asks for them.

Thank you for supporting our sponsors, who keep this newsletter free.

Interested in sponsoring these emails? See our partnership options here.

For a while, AI’s impact on the SEO world looked like an ability to pump out thousands of blog articles a day and rank for every possible search term. But, that plan assumed buyers would keep searching the way they always had. In reality, more of them now describe their problem to AI, discuss the answer in a chat, and come out the other side holding a shortlist of products. You have no idea how, or even if, you were described along the way.

I ran my own agency through an AI visibility report this month. It told me that when AI is asked about Pistachio directly, the models recite my own messaging almost word for word, an owned media engine, founder-led content, the missing middle. But when the same models are asked non-branded, problem-stage questions, we weren’t mentioned at all.

Traditional search rewarded the exact phrase someone typed, so the old playbook worked from the bottom of the funnel up. You wrote the comparison page for "best email software", then the category page like “10 reasons you need email software”, then something further upstream if budget and time allowed.

Andersen frames the new search journey as problem recognition, then solution research, then vendor evaluation, which he said is "much more aligned with kind of a sales methodology". It matches how people work through problems, but now they talk to AI at every step. They arrive with a problem, and the conversation does the narrowing.

Researchers classified consumer ChatGPT messages sent between May 2024 and June 2025, and found about half of them were people seeking information to inform a decision. That’s consumer use across every topic rather than buyers alone, and it’s still the single biggest thing people do with AI.

So the questions that decide your competitor shortlist get asked well before someone asks those bottom of funnel questions, and nothing in a list of keywords will tell you what they are. Andersen says "on an average query it fetches about 50 sources, it reads about five of them, and the answer lands on about three to four options". His own example was buying headphones, where he used to work through reviews and feature comparisons himself, but "nowadays increasingly people just kinda ask AI and it goes, ‘Yeah, this is the best’ and they go, ‘Cool, good enough’".

Andersen splits the new frontier of AI search into two battles. Showing up at all, and how you are described when you do. It’s the same as my brand strategy work, getting attention is the first challenge, then it’s the perceptions that are built from it.

My own report was split along that line. Ask about the agency and AI sings from our song-sheet about founder-led content and owned media. Ask a broader question and we don’t appear at all.

The models could read our claims but had nothing outside our own site to validate against, and they said as much, that our public evidence was thinner than the agencies they named instead. There's case studies on our site, but that’s our own domain with nothing external backing it up.

Luckily, both battles are cheap and easy to test. Paste your site into ChatGPT and see whether it can read it at all, because AI doesn’t crawl your site the way Google does. Set yourself up in Google Search Console and Bing Webmaster, then ask two different models a buyer question that doesn’t mention your name, and compare the answers.

Most of the advice flooding in about AI search visibility still treats it like a ranking system. Track the prompts, publish the keyword pages, and get your name to the top of the sources list.

One SEO freelancer tested that approach this year, running 100 B2B "best category" queries through Google's AI Overviews across three checks from April to June. When a brand's own "best of" listicle was cited as a source, the brand itself was left out of the AI’s recommendation almost 70% of the time. Google read the page, took the useful information, and used it to recommend somebody else.

Andersen sees the same pattern from the other side, saying "I see brands with fifty different listicle pages", and he sees them increasingly knocked down by Google spam updates. The models are getting better at flagging claims they can't verify, which turns a page of self-praise into a genuine liability.

Google ran its third spam update of the year in August, and Andersen was proud to say his clients' rankings all went up, while others posted graphs heading straight down.

Volume carries a second cost, because every page you publish is another page that can contradict the others, and AI reads all of them. "If you've tried to manage a blog or a website over like four years, you barely understand what was going on back then", Andersen said. He worked with a large finance comparison site and found different pages told buyers different stories. Then again at a company launching a new product line, where AI kept describing them by the old one because that was what most of the pages still said. Where old school keyword tactics ranked each page separately for the information it held, now everything you publish becomes part of one picture AI uses to assess you.

"Most of the effort is just surfacing all the information that you internally already believe", Andersen said, beginning with the questions buyers already ask you. Your sales or support teams answer them every day, in private.

It’s those unique insights that hold all the gold. Google's spam policies go after scaled content abuse, with Andersen saying "many pages are generated for the primary purpose of manipulating search rankings and not helping users". He gave me a simple question to ask. "Could this be on any company in your space's blog and it wouldn't make a difference?".

The pages themselves can be plain. "Saying “I'm the best in the world” is not useful", Andersen said. "Saying “we solve for this person, they typically get this return or this result” is much better".

The page he rates above every other is a guide to the problem itself, setting out which kind of solution suits which kind of person. His example is if you have an injury, the right answer could be self-medicating, a GP, or the emergency room. Different solutions are best suited to different contexts, so helping AI be your sales rep means clearly defining who you’re a fit for, and routing those you’re not elsewhere. A guide that sends the wrong buyer elsewhere tells AI exactly who the right one is.

"It is much easier to change AI's mind than it is consumers", Andersen said, and that is the half of this you can actually work on. The rep will repeat what you give it, in the words you gave it, to people who will never see your homepage. But in the end, a person still makes the decision (at least for now…).

The brands AI recommends are the ones that make it clear and easy to understand who they’re for, and have verifiable proof somewhere other than their own site. Keyword tactics built for search ranking wear out with every update.

If you enjoyed this post or know someone who may find it useful, please share it with them and encourage them to subscribe: brandchemistry.co/p/why-ai-search-is-a-brand-problem