Start from the question, not the website
Most companies audit their website by reading it. We start somewhere more useful: the questions buyers actually ask AI. By mapping realistic buyer prompts against what a company’s public information can actually support, the specific gaps become obvious — and so does the order in which to fix them.
Illustrative example: the prompts
Consider a sterile injectable CDMO. Realistic buyer prompts an AI model might receive include: “Recommend a CDMO with sterile injectable and lyophilisation capability approved for the EU,” or “Which CDMOs have tech-transfer experience for sterile injectables and a strong regulatory record?”
Each prompt is really a checklist of facts the buyer is filtering on: sterile injectable capability, lyophilisation, EU approval, tech-transfer experience, regulatory track record.
Illustrative example: the gaps
Mapping those prompts against the company’s public information reveals where AI would come up short. Perhaps sterile injectable capability is stated clearly, but lyophilisation is never mentioned; EU approval is implied but not named; tech-transfer experience exists but appears nowhere on the site. Every unmet item is a prompt the company would silently lose — not for lack of capability, but for lack of a visible, extractable fact.
The output is a prioritised list: the specific facts to add or clarify, ranked by how often they appear in the buyer prompts that matter most.
Why this matters commercially
Working backwards from real buyer prompts keeps AI Discovery focused on commercial outcomes rather than generic “best practice.” It ensures effort goes to the gaps that are actually costing shortlist places — the questions real buyers ask — rather than to cosmetic changes that do not affect whether AI recommends you.