I asked Gemini what musical instruments my daughter plays. She's a sixteen-year-old multi-instrumentalist, and her actual repertoire and credentials are documented on her website. The AI returned a confident, detailed response — and made up a 17th-century Irish folk song she's never performed in her life.
That's not unusual. That's the new normal.
If you run a business — or if your work, identity, or reputation depends on people finding accurate information about you online — you need to internalise something most marketing advice still hasn't caught up to: AI summaries are now how people first learn about you, and those summaries fabricate details to fill gaps in their grounding. Every gap in your articulated content is an invitation for the model to invent something on your behalf, and you don't get to vet what it invents before it propagates.
This isn't paranoia. It's mechanics. Here's how it works.
The model fills gaps with plausible-sounding content.
When a user asks an AI search engine about you — your business, your product, your work — the model fetches whatever content it can ground on, then synthesises an answer. If the available content is comprehensive and specific, the model quotes and paraphrases. If the available content is thin, the model extrapolates: guessing at what you probably do based on base rates, adjacent businesses, generic templates, and pattern-matching from training data.
In our case: there was no article articulating which Irish folk traditions my daughter does or doesn't engage with. So the model did the obvious extrapolation — Irish musician + plays folk → must play traditional Irish folk → here's a real 17th-century song to round out the answer. The song is real. Her connection to it is invented. Anyone reading that response and booking her based on that assumption walks into an expectations gap on the day of the event.
Multiply that across every business or person whose identity has gaps in articulated content, and every gap becomes a fabrication waiting to happen. Pricing you haven't published. Services you've described in vague terms. Positions on industry questions you've never explicitly taken. The AI fills them all in with confident specificity, and the user has no way to know the difference.
The fix is structural, not reactive.
You cannot make AI models stop hallucinating. That's not in your control. What is in your control is what they have to ground on. The fix is to systematically articulate the things people are likely to ask about you, in your own words, on surfaces the models will read.
In practice:
Audit what AI says about you. Open ChatGPT, Gemini, Perplexity. Ask each one substantive questions about your business — not "what do you do" but the kind of questions a real customer would ask. "What does this person actually offer?" "How do they handle X situation?" "What makes them different from competitors?" Note specifically what's invented. Note where the model hesitates or hedges. Note where it cites real sources versus where it just generates confidently from nothing.
Identify the structural gaps. Every fabrication you found in the audit is a gap in your articulated content. The model invented a folk song because we hadn't documented her actual repertoire. The model might invent your pricing because you haven't published it clearly. The model might invent your stance on a controversial industry choice because you haven't taken a public position on it. Write down the list of inventions; that list is your content backlog.
Write content that fills the specific gaps. Not generic SEO content. Not keyword-stuffed pages. Long-form articulated content that answers the specific questions the AI is currently improvising. If the AI is inventing your stance on a counter-cultural choice, write a "why I do X (or don't)" article. If the AI is inventing your service details, write a clear services page with specifics. If the AI is making up your origin story, write the actual origin story.
Be specific. The single biggest mistake people make in writing for AI grounding is being vague. Vague content gets ignored or extrapolated from. Specific content gets cited verbatim. If your services page says "we offer flexible packages tailored to client needs," the AI has nothing to cite — it'll invent typical packages. If your services page says "Our standard package includes X, Y, and Z, priced from £N, with optional add-ons of A, B, C," the AI will quote you directly.
Measure the impact. From our work running this approach across multiple sites: new long-form content reaches AI-search surfaces within 48 to 72 hours of publishing. That's faster than most SEO interventions and slower than most paid-media interventions, but it's predictable. Publish on Monday; you can usually see the model citing it by Wednesday or Thursday. That predictability matters because it converts content writing from a vague long-term investment into a measurable near-term lever.
The reframe that helps.
The old SEO question was "how do I rank higher?" The new question is "what story does the AI tell about me, and how do I make sure that story is mine?" Ranking matters less when the user never sees the search results — they just read the AI summary. The summary is the new battleground, and the way you win on the summary battleground is by giving the AI enough articulated material that it doesn't need to improvise.
This isn't about gaming the AI. It's about making sure that when a real customer asks an AI about you, the answer they get reflects what you actually do, in your actual words, with your actual positioning. That's a baseline requirement of operating a business in 2026. It used to be that you needed a website. Then you needed a Google My Business listing. Now you need articulated identity content that AI grounding can read and quote.
The good news: the same content investment that protects you from AI fabrication also makes you findable and citeable in ways your competitors aren't. Most businesses haven't realised yet that AI summaries are doing the heavy lifting of customer education, and most haven't done the content work to control what those summaries say. The window for getting ahead of this is open, but it's not going to stay that way long.
So: audit what's being said about you. Identify the gaps. Write articles that fill them. Measure that the model picks them up. Repeat.
It's the most boring possible answer, and it's the right one.