What AI Says About Your Brand (And How to Fix It)
AI answers about your brand are shaped by third-party sources you don't control. Here's how to audit yours and start influencing the public record.
There's a version of your brand on the internet that you didn't write, didn't approve, and probably haven't read. It's the version AI assistants quote when someone asks them about you. For most small businesses and indie creators in 2026, that version is a mash-up of an old Trustpilot review, a Reddit thread from three years ago, and the comparison page on a competitor's blog. The good news: you can change it. The less good news: you have to know it exists first.
What is the "public record" in 2026?
The public record is the set of things AI systems can find and quote about your brand when asked. It's bigger than your search-engine result. It includes the things you publish (your website, blog, product pages), the things others publish about you (reviews on Trustpilot, Google reviews, Yelp, threads on Reddit, posts on LinkedIn, comparison roundups on competitor blogs), and the things AI training datasets happened to absorb when the model was last updated.
Most small site owners think only about the first category. AI assistants weight all three. When you ask Perplexity "is <your business> any good?", the answer is built from whatever surfaces it can retrieve right now, plus whatever the underlying model already learned. You wrote 5% of those inputs. The internet wrote the other 95%.
Why is this different from old-school SEO?
Search engine optimisation (the long-running discipline of getting your pages to rank on Google) is about being found when someone clicks through to a website. The public-record problem is different: the user often never clicks through at all. Google's AI Overviews summarise the answer in-page. Perplexity reads ten sources and synthesises one paragraph. ChatGPT (when web-search is on) quotes a couple of lines and moves on. That whole conversation can happen without anyone ever visiting your site.
So the rules shift. It's not enough to have a beautiful homepage and a well-tuned title tag. The places AI tools read have to say the right things about you. If the loudest signal Reddit gives off about your product is a two-year-old complaint thread, that's the line the AI will quote, no matter how good your homepage looks today.
How do I audit my own public record?
Spend 30 minutes with the AI tools your customers actually use. Treat it as research, not vanity searching.
Open ChatGPT (with browsing on), Claude, Perplexity, Google Gemini, and Google's regular search (with AI Overviews enabled). Ask each one the same four prompts about your brand or product. Write the answers down verbatim. You'll see patterns within ten minutes.
The blunt one
"Tell me about <your brand>." What's the first thing it leads with? Is it accurate? Is it from this decade?
The comparison one
"What are the alternatives to <your product>?" Are you in the list? Where? What does it say about your weaknesses?
The recommendation one
"Should I use <your brand> for <your category>?" Does the answer recommend you or steer toward someone else? Why?
The category one
"What's the best <your category> for <your target customer>?" Does your name appear? In what context?
You will find one of three things. Either the AIs don't mention you at all (you have no public record yet), they mention you with broadly correct framing (the record is fine, work on visibility), or they repeat one specific outdated or unfair claim across every prompt. That last case is the one to chase, because it usually means a single high-authority source (one Reddit thread, one G2 review, one old TechCrunch article) is dominating the signal.
What can I actually change?
You can't edit Reddit. You can't take down a competitor's comparison page. What you can do is publish enough better-sourced material that the loud signal stops being the loudest one. There are four levers, ranked roughly by leverage.
Your own site. Publish the answer to the question the AI is currently getting wrong, in plain language, with a clear page URL the AI can find. If the wrong line is "<your product> doesn't support Linux" and you launched Linux support last year, write a single post titled "Yes, <your product> works on Linux" with the install steps. It's a small thing. It often shifts the AI answer within weeks.
Structured signals. Add JSON-LD schema to your product pages so the AI can extract facts cleanly (price, ratings, what category you're in, what you're not). Add an llms.txt file at your site root listing the URLs you most want AI assistants to read. Both are five-minute changes that materially affect what gets quoted.
Third-party surfaces. When customers ask if they can leave a review, send them to Trustpilot or Google. When journalists or bloggers cover your category, make sure they have your current product page open. Aged third-party signals beat young first-party ones in most AI pipelines.
The places competitors aren't. Pick one industry-specific community (a subreddit, a Slack group, a forum) and turn up consistently with useful answers, not pitches. Six months of helpful posts under a real human identity beats one promotional thread.
A 30-minute starter audit
If you read everything above and stalled, this is the version to start with. Set a timer, open a Google Doc.
Five minutes: the four prompts
Run the four audit prompts (above) against ChatGPT, Perplexity and Google AI Overviews. Paste each answer into the doc.
Five minutes: spot the pattern
Read the answers back-to-back. Highlight any phrase, claim or comparison that appears in two or more of them. That's your loud signal.
Ten minutes: find the source
For each loud signal, ask the AI "where did that come from?" Most will cite. Open the source. Decide whether it's accurate today.
Five minutes: pick one thing to publish
If a loud signal is wrong or outdated, write down the single page on your site you'd need to publish (or update) to correct it. Don't write the page now; just name it.
Five minutes: schedule it
Put the page on next week's calendar. Re-run the audit four weeks after publishing. Compare.
When should you not bother?
If your business has fewer than ten customers, the AI public record isn't a problem worth solving yet. AI assistants need data to retrieve, and a brand-new brand simply doesn't have any. Use the first 6-12 months to publish your own pages, get a handful of honest reviews on Trustpilot or Google, and post under your own name in two or three relevant communities. The audit becomes useful once there's something to audit.
Likewise, if you're a hobbyist with no commercial intent, the cost of fixing the public record is rarely worth the time. The audit is genuinely interesting (you'll learn what AIs think of you), but the playbook below is calibrated for people whose income depends on the answer.
20 Actually Useful Things to Ask ChatGPT
Free AI Tools You Should Be Using in 2026
