Do the exercise before you finish this paragraph. Open ChatGPT, or Gemini, or Perplexity, and ask three questions about your own company: what does it do, what is it best at, and what should a buyer know before choosing you over the obvious competitor. Read the answers slowly, the way a prospect would. Somewhere in there is a sentence that made you wince. A competitor described with more warmth than you would like. A product you killed in 2023 living on in the present tense. A pricing claim that was never true, delivered with the flat confidence of a court record.
That is your AI reputation. It is assembled from your website, your reviews, a Wikipedia stub, old press coverage, forum threads you have never read, partner pages that copied your copy in 2021, and whatever else the models absorbed before their training cutoffs. It gets recited, fluently and without hedging, to buyers who will never visit your site to check. And here is the organizational fact that should bother you more than any single wrong sentence: nobody in your company owns it. PR owns journalists. SEO owns rankings. Brand owns the deck with the pyramid on it. The answer a machine gives about you at the exact moment of consideration falls in the gap between all three job descriptions, and gaps do not fix themselves.
The scale stopped being hypothetical this year. Assistants now sit inside the buying process the way search did in 2005, early enough that most of your competitors are asleep, late enough that the behavior is measurable in any pipeline review. When I run a structured audit of a brand's AI reputation, the battery is deliberately unglamorous: category questions, comparison questions, objection questions, and the brutal one, what are the downsides of working with this company. Twenty questions, four engines, scored on three axes: factual accuracy, sentiment, and citation share, meaning whose pages the answers actually lean on.
The pattern across dozens of these audits is remarkably stable. The answers run roughly seventy percent accurate. The errors cluster around whatever changed most recently, pricing, packaging, leadership, positioning, because the record trails the reality. The downsides question is where reputations go to be ambushed: I have watched engines answer it from a single aggrieved review thread years old, elevated to the status of consensus because nothing else on the record addressed the topic at all. And the framing of every answer borrows disproportionately from whoever published the clearest, most structured explanation of the category, which is frequently not the market leader. Somebody is supplying the language the machines use to describe your market. If it is not you, it is someone selling against you.
That last finding is the entire opportunity, and it is worth stating as a rule: these systems do not reward size, they reward legibility. A serious generative engine optimization program is, stripped of vendor noise, a legibility program. Claims stated in complete sentences a machine can quote without reconstruction. Structured data that parses and matches what a human can see. Entity facts, name, founding, leadership, locations, kept identical everywhere your company appears. Content that answers the questions buyers actually ask instead of the ones your homepage wishes they would. The brands winning citation share right now are the ones a machine can repeat without hedging, and this site is built as the working proof, down to the machine-readable edition of everything on it.
So who should own the AI reputation? My answer after building these programs: whoever owns organic demand, with a dotted line to communications, and a standing quarterly audit on the calendar the way media mix reviews used to be. The first audit takes one afternoon. It produces the most clarifying document a leadership team has seen in years, because it is the first time anyone has read the company's story the way a machine tells it, and the fix list writes itself: retire the pages that lie, answer the questions that sit vacant, put a named human behind every claim, and correct the entity record everywhere it drifts.
The uncomfortable close is also the motivating one. Your AI reputation is being written right now, mostly by your least-maintained pages and by other people's opinions of you, and the machines rereading the record daily will keep reciting whatever they find. That is the threat and the entire mechanism of repair in one sentence. You cannot retrain their models. You can clean the record they read, and the record is yours. Assign the owner this quarter, run the twenty questions, and start correcting the file while your competitors still think this is a 2027 problem.