Generative AI
What it does well
- A first draft from your notes: a news item, a letter, a grant application
- The summary of a long document, a call for projects, a regulation
- Answers to questions whose answer is in your documents, through an assistant linked to the website
- A translation to proofread, a rewording in plain language
- Sorting: classifying emails, extracting dates and amounts from a table of requests
What to watch
- It can invent: proofread everything that contains a figure, a name, a date, a legal reference
- Some free tools use your conversations for training: no personal data of beneficiaries or members in them
- It does not replace judgement nor relationships: a letter to a family in difficulty is proofread, signed and owned
- It standardises: an entirely generated text is recognisable, and your publics recognise your voice
Where to start
- Choose a use case without personal data: the website's news, summaries of calls for projects
- Give the tool your own texts as material (annual report, website pages) rather than asking it to invent
- Write a one-page usage rule for the team: what goes in, what does not, who proofreads
- Measure after a month: time saved, errors caught, what the team thinks
Concrete examples
- A social grocery summarises each week the new calls for projects of its region in ten lines, and reads only two in full
- A youth centre turns the minutes of a meeting into a news item for the website, proofread in five minutes
- A help service translates its opening hours and access conditions into four languages, then has volunteers proofread them
- A federation answers its members' questions about its rules through an assistant that only reads the rules
In Kern4Good
The programme offers a half-day discovery session for teams and, when useful, an assistant linked to the association's website, built on KERN-IT's assistant platform, with clear rules on data: the AI for non-profits theme describes the approach. These services are quoted at cost, on the Pricing page. The programme's websites are also written to be read by AI assistants: short answers, sourced facts, an llms.txt file.
Further reading
Is it GDPR compliant?
Yes if the tool and its settings allow it: no personal data in exchanges with a consumer tool, a contractual framework for a professional tool. The programme helps you choose.
How much does it cost?
Consumer tools have a free version and subscriptions at a few dozen euros a month. An assistant linked to your website is quoted at cost.
Can an AI write the whole website?
A first draft, yes, and the programme's launch website relies on it. But the facts (hours, conditions, figures) come from you, and so does the proofreading. An entirely generated website without proofreading ends up containing errors your publics see before you do.
What is an assistant on our website for?
To answer day and night the questions whose answer is in your pages: who can benefit, how to register, where and when. It only answers from your documents, with the rules you set, and refers to a human for the rest.