Tenderhooks uses AI for one job: reading each research or consulting tender and writing a short plain-English summary, a checklist of what the buyer asks bidders for, and sub-trade labels from a fixed list. Which tenders reach you, and every fact and figure, come from ordinary code.
What the AI does
- It reads the public text of a tender notice (its title, buyer and description) and writes a summary of at most two sentences and a checklist of what the notice asks for.
- Before anyone sees it, ordinary code checks every summary and checklist. If one holds a date, an amount of money, a link, an email address, a phone number, a name with a title (such as "Dr Jones") or a number the notice doesn't contain, it is not shown, and you see the notice's own words instead.
- It suggests sub-trade labels from a fixed list. Matching itself uses the notice's official codes and words, in code.
What it never does
- It never supplies facts you act on: deadlines, values, buyers and dates always come from the official notice.
- It never decides "Worth a bid?": those figures are computed in code from official award data.
- It never acts: it has no tools, sends nothing and changes nothing.
- It never sees personal data about you, your team or your searches. It gets only the public notice text, with contact details removed.
- Your data is not used to train any model.
How you can tell
Every summary sits under a visible "AI summary" label, next to the link to the official notice. The tender page and the emails also say that AI summaries are not official, may contain errors and should be checked against the official notice. Where there is no summary, the notice's own words are labelled "Official description". Web pages and emails also mark AI summaries in a machine-readable way.
The model
Summaries are written by Claude, made by Anthropic, through its API under Anthropic's commercial terms, which do not allow our inputs to be used for training. If a summary is wrong, harmful or inappropriate, tell us and we will correct or remove it.