Six questions to put to an AI vendor
Six questions to send in writing before the demo, one for each layer of TRACES, with what a usable answer looks like and what an evasive one sounds like.
ReadInsights
Writing on AI you can account for.
Articles, whitepapers and the occasional note. On sources, sign-off, and what holds up in the thinking professions.
Voor wie
Education and learning materialsConsulting and strategyPublishers and editorial teamsResearch and think tanksBBEbuddyBookaflowValoraEducation and learning materialsConsulting and strategyPublishers and editorial teamsResearch and think tanksBBEbuddyBookaflowValora
Sorting a workflow: which steps AI may do
"Can we use AI?" is a question you answer step by step, not for the work as a whole. Three questions and three bins for sorting a workflow.
ReadAn AI work policy that fits on one page
A policy nobody reads decides nothing. Six blocks on one A4 page, with the question each block answers, and a worked example.
ReadWhat to record when you work with AI
Five fields that make a step with AI traceable, even a year later. What belongs in them, what doesn't, and how the trail builds itself.
ReadWhich changes need a substantive judgement
Not every change to a text deserves the same attention. A three-bin test for deciding which changes need a subject-matter judgement and which can simply be read along.
ReadTesting a knowledge assistant with questions that have no answer
An assistant that answers good questions well proves little. A test set of twenty questions, half with no answer in your sources, shows whether it can also say it doesn't know.
ReadSetting up an AI pilot you can decide on afterwards
Many pilots end with "it went quite well" and then nothing happens. A pilot card with six fields, filled in beforehand, means you can choose between continuing, adjusting and stopping when it's over.
ReadWhat a school board asks before teachers get an AI tool
A board that has to say yes or no to an AI tool gets more from six written questions than from a demo. Each question comes with the document the school itself should have ready.
ReadLetting teachers use AI without giving away student data
An agreement that teachers won't put student data into a chatbot gets broken in a busy week. What works instead: swap the data out before the prompt, and set up the environment so it's also technically hard to do.
ReadAssessing a report that was made with AI
A report full of AI errors looks just as polished as a good one. A sample check on the sources shows the difference, and takes an afternoon.
ReadThe AI paragraph in a bid or tender
Nearly every AI paragraph in a proposal promises responsible use and says nothing else. A paragraph of four sentences, one per SEAL phase, does say something.
ReadChecking an AI summary for shifted meaning
A summary can be right in every detail and still claim more than its source. A three-point check for newsrooms that let AI summarise.
ReadCarrying a correction through everything built on the same source
One error in a source soon sits in five or ten pieces of content. Fix it in one place and the rest stays wrong. A list you build at publication solves that.
ReadRemoving invented sources from a literature review before delivery
An AI literature review contains references that look right and don't exist. A three-step check, per reference, and a log you can hand over with the work.
ReadChecking the step from finding to recommendation
The findings are right, but the recommendation goes further than they carry. A support table shows, for each recommendation, where the leap is and who gets to make it.
ReadThe Working Session, € 399.
Ninety minutes with the core team along your own workflow. Where it may go faster, and where credibility is at stake. You get one A4 with three observations and one recommendation. The amount is credited against a scan within three months.
The Working SessionIntro call, free.
A half-hour conversation about where your hours go. We won't chase you afterwards. You'll know whether this is for you, and so will we.
Book an intro callThe Accountability Scan, fixed price.
We go through your AI use and knowledge work. What is demonstrable, where the gaps are, where the routine work sits and what it costs. You get a report in plain language, with an order of attack and a cost estimate. The report is yours, whether we continue together or not.
The AI Accountability ScanAfter that: setting up a tool, training a team, or building a system. And when it pays off, a standing partnership in which we review the AI use at set intervals and you get a written report each time. What that involves is in the offer.See the offer