Most companies that start worrying about AI search jump straight to the competitive question: why does ChatGPT recommend our competitor instead of us? It's the right worry, but the wrong starting point. If an AI assistant can't describe your business accurately, it has no good reason to recommend it.
At SEO Week 2026, Wil Reynolds shared a maturity model developed by Alisa Scharf at Seer Interactive that puts this in order. You earn your way from accuracy, to being associated with the right specialisms, to appearing in long-tail category answers, and only then to the broad, competitive recommendations everyone wants.
1. Write down the facts you want AI to know
Start with a brand fact sheet: what you do, who you serve, where you operate, what makes you different, who leads the company, and anything that has recently changed. Seer's approach goes deep here, cataloguing dozens of attributes. For a small or mid-sized B2B company, twenty to thirty well-chosen facts is a solid start.
Then make those facts consistent everywhere they live: your About page, service pages, structured data, LinkedIn, directory listings, and review profiles. AI systems learn from many sources, and contradictions between them are one of the most common causes of wrong answers.
2. Build a fixed set of test prompts
- Direct questions: where is the company based, what does it offer, who does it serve?
- Open questions: tell me about this company.
- Comparisons: how does it compare with a named competitor?
- Category questions: who are good options for this service in this industry?
Keep the set fixed so results are comparable over time, and run it across several assistants. Answers vary from one run to the next, so look for patterns rather than single responses.
3. Score accuracy before share of voice
For each answer, score whether the facts are right. This accuracy score is your leading indicator. Share of voice in competitive answers is a lagging one: it rarely improves while the basics are still wrong.
4. Trace every wrong answer to its source
When an answer is wrong, check which sources it cites, then search for those claims yourself. Often the problem is an outdated directory listing, an old press release, or a third-party page repeating a mistake. Fix it at the source, not just on your own site.
Ray Martinez of Archer Education added a useful diagnostic at the same event: separate visibility problems from comprehension problems. If AI tools can't find your pages, that's a crawling and authority issue. If they find them but misread them, it's usually structure: heavy JavaScript, unclear headings, or key facts buried in images and PDFs.
If AI can't describe you accurately, it has no reason to recommend you.
Accuracy work is slow to show up because models learn from historical snapshots. The sooner correct, consistent information exists, the sooner it begins shaping future answers.