--- type: Article title: Your Brand Has an AI Reputation Now description: >- AI assistants don't list sources about your brand. They write a description of it. Here's how that description gets built, and a 20-question audit to see yours. resource: 'https://adityarsingh.in/blog/your-brand-has-an-ai-reputation/' tags: - AI Search timestamp: '2026-10-01T00:00:00.000Z' --- # Your Brand Has an AI Reputation Now Search a company's name on Google. Then ask an AI assistant what that company is known for. For most of the last twenty years, those were roughly the same exercise. They aren't anymore. Take a fictional company. Call it Halden, a mid-sized HR software vendor. Google gives you a familiar stack: halden.com, a Wikipedia page, a review site listing, a couple of news stories, a LinkedIn page. Ten links, and you decide what to make of them. The assistant gives you something else. It says Halden is "a payroll and HR platform popular with mid-market companies, known for its compliance features, though some users find onboarding slow." That second answer has no list in it. It's a description. Something already read the sources and decided what they add up to. I've been staring at that gap for months, mostly while doing GEO work. In [Index Yourself](/blog/index-yourself-what-llms-know-about-you.md) I looked at what a model knows about a person. This post asks the same question about a brand. I've started calling the answer the brand's AI reputation. ## How AI turns sources into a description Companies have always had reputations. For a long time, they also had a home base for them. The homepage, the product pages, press releases, ads, executive interviews. Journalists, customers and competitors always had a voice too, but the brand owned the canonical version of its own story. Search added a layer on top. Now the question was also what the web said about you. But search still handed the work back to the person. They typed a query, opened a few tabs and formed their own opinion. AI assistants do that last step for them. Picture a hundred pages that say Halden is good for compliance. Seven say it's expensive. Three say implementation drags. Nobody reads all 110. The assistant reads what it retrieves and writes one sentence: a strong compliance platform, though pricing and setup can put off smaller teams. That sentence is reputation synthesis. The model found the information, then decided which parts mattered and how much weight each one got. The weighting used to happen in a buyer's head. Now it happens before the buyer shows up. ## Most of your AI reputation lives on other people's sites A brand controls its own website. It doesn't control review sites, Reddit threads, YouTube walkthroughs, Wikipedia, analyst reports, news coverage, comparison pages, partner listings or a former employee's LinkedIn post. AI systems read all of it. Two recent studies put rough numbers on this, and they measure different things. [Slate's study of brand-name searches](https://slatehq.com/blog/ai-overviews-brand-searches), published October 1, 2026, ran 100 brand names through Google. Only 29% of the 652 sources cited in the AI Overviews pointed to the brand's own site. Thirteen brands supplied none of their own sources. [Agenzy's citation study](https://www.agenzy.lt/blog/which-sources-ai-engines-cite-2026) looked at 394,180 AI answers to buyer questions across ChatGPT, Google AI Overviews, Gemini and Perplexity. Brands' own websites earned 3.7% of citations. Their named competitors earned 16.9%. I wouldn't treat either number as universal. The first is one US scraper over two days. The second is one agency's client and pitch dataset, three quarters of it businesses in Lithuania, and it measures category questions, where you'd expect a brand's own site to show up less. Different methods, different markets. But they point the same way. Even when someone types your exact name, most of what the assistant says about you comes from somewhere else. ## The three layers of an AI reputation When I read an AI answer about a brand, I've started splitting it into three layers. They fail in different ways, so it helps to look at them separately. ### Identity: who are you? This is the factual base. What the company does, what it sells, who runs it, which market it's in, and whether the model has the right company at all. A brand with a common-word name can lose this layer entirely. It's the same problem I ran into in [Index Yourself](/blog/index-yourself-what-llms-know-about-you.md), just at company scale. ### Perception: what are you known for? "Halden sells payroll software" is a fact. "Halden is strong on compliance" is a judgment, built from how often and how confidently the web repeats it. Two companies with identical products can end up known for completely different things. ### Framing: how are you described? This is the layer I find most interesting. "A leading HR platform" and "an older HR platform" can describe the same company. Both might be defensible. They leave a buyer with very different pictures. I touched on this in [Cited vs Ranked](/blog/cited-vs-ranked-discovery-economics.md). A ranked list never described you. An answer describes you every time it mentions you, in its own voice, with the borrowed authority of something that sounds neutral. ## When the AI gets it wrong without hallucinating Most conversations about AI and brands start with hallucination. What if the model makes something up? That happens. The more common problem I see is quieter. The model reads a messy web and resolves the mess badly. Say Halden changed CEOs last year. The website and LinkedIn name the new one. An old press release, a startup database and a few interview pages still name the old one. The assistant answers with the old name. Calling that a hallucination misses what happened. Every source it used exists. The model weighed five pages against two and picked the majority. Your information was inconsistent, and the answer reflected that. Competitors can shape the picture the same way, without touching your site. Slate's study found brands whose AI Overview cited a rival's "what is [brand]" article, or a comparison page written by a competitor. If a rival's page about you ranks, it can become part of how you're described. Same with a four-year-old forum thread, or a review from before you fixed the thing it complained about. None of this requires the brand to do anything wrong. The outside web tells a different story than the homepage does. I now think of a brand as a distributed record, scattered across hundreds of pages, that a model tries to stitch into one entity every time someone asks. ## How to audit your AI reputation in 20 questions In Cited vs Ranked I suggested running twenty buyer questions about your category. This audit points the same idea back at the brand itself. Asking "what is [Brand]?" once tells you almost nothing. Twenty questions across four groups start to show you the whole picture. ### Identity 1. What is [Brand]? 2. What does [Brand] do? 3. Who is [Brand] for? 4. What products or services does [Brand] offer? 5. Who runs [Brand]? ### Perception 6. What is [Brand] best known for? 7. What are [Brand]'s main strengths? 8. What are [Brand]'s main weaknesses? 9. Is [Brand] a good fit for [your core use case]? 10. How has [Brand] changed in the last year or two? ### Comparison 11. Who are [Brand]'s main competitors? 12. [Brand] vs [Competitor] for [use case]? 13. What are the best alternatives to [Brand]? 14. What's the best [category] option for [your target customer]? 15. Why would someone pick [Competitor] over [Brand]? ### Trust 16. Is [Brand] reliable? 17. What do customers complain about with [Brand]? 18. Is [Brand] worth the price? 19. Is [Brand] safe to use with sensitive data? 20. Would you recommend [Brand]? Question 14 is the only one that doesn't name you. Keep it anyway. It shows whether you exist in the answer when the buyer hasn't heard of you yet. Run all twenty in ChatGPT, Gemini, Perplexity and Google AI Overviews. Use a logged-out or clean session so your own history doesn't color the answer. Then log each answer in one row. | Question | Engine | Mentioned? | Cited? | Sources used | One-line description | Framing | Competitors named | Conflicts with your facts | | --- | --- | --- | --- | --- | --- | --- | --- | --- | | 8. Main weaknesses | Perplexity | Yes | No | Review site, forum thread | "Slow onboarding, dated UI" | Negative | 2 | UI redesign shipped, not reflected | The last column is the one that turns this into work. Every conflict points to a specific page somewhere that's telling the older or wrong story. ## Ask the same question ten times This is the experiment I'd run first if I could only run one. Pick question 6, "What is [Brand] best known for?", and ask it ten times in fresh sessions. Write down the first descriptor each answer uses. For a fictional company like Halden, you might get payroll four times, HR software three times, compliance twice and workforce management once. Same company, same question, four different identities. Then ask a follow-up: why did you describe it that way? The sources it points to usually explain the drift. The variation is real at the most basic level, too. In Slate's study, 16 of 44 repeated brand searches flipped between showing an AI Overview and showing none. Slate only reran the brands that had shown no Overview or a glitch the first time, so that's the shakiest part of the sample. Still, if the decision to answer at all changes between identical searches, the wording inside the answer will too. That's why I think stability deserves its own number. A brand that gets described the same way eight times out of ten has a settled AI reputation. A brand that gets four different descriptions has a reputation the web hasn't agreed on yet. The second case is where the work is. ## What to keep watching after the audit The audit is a snapshot. For ongoing tracking I'd skip the giant checklist and watch four things. They're the three layers again, with perception split in two, because the category you get filed under and the story told about you drift separately. **Entity accuracy.** Does the model know who you are, and are the basic facts current? This is the easiest layer to fix and the most embarrassing one to get wrong. **Category association.** Which shelf does it put you on? A company that sees itself as a platform but keeps getting filed as a point tool has a positioning problem the AI is just reporting back. **Narrative.** What are you known for, and how stable is that across runs? The ten-times test gives you the number. **Framing.** How does the answer describe your strengths, weaknesses and differences from competitors? Watch the adjectives. "Popular" and "legacy" are both one word, and they do very different jobs. The shorthand I use is AI reputation = identity + association + narrative + framing. The first two are mostly facts you can correct at the source. The last two take longer, because they come from what the rest of the web repeats about you. ## Reputation is machine-readable now Your brand always had a reputation. What changed is that a machine now summarizes it for your customers, in one paragraph, before they've opened a single tab. The web writes that paragraph, and your marketing team is one voice among hundreds. Your homepage is in it. So is a competitor's comparison page, a forum thread from four years ago and a video review you've never watched. Getting found by AI was the first problem. The one I'm more interested in now is what happens after. When a model reads everything the web says about you and writes it down, does it describe the company you think you are? Run the twenty questions and find out. One engine takes about an hour, all four an afternoon, and you'll probably learn something about your brand you didn't know. --- ## Sources - Slate, [What Google's AI Overviews Say About Your Brand: 100 Brand Searches Analyzed](https://slatehq.com/blog/ai-overviews-brand-searches), October 1, 2026 - Agenzy, [Which sources do AI engines cite? 2,283,923 citations across ChatGPT, Google AI Overviews and Perplexity](https://www.agenzy.lt/blog/which-sources-ai-engines-cite-2026), September 8, 2026