How to Build an AI Content Repurposing Workflow for UK SMEs

What to take away

Start from the reader's job, not the tool

Before opening any AI assistant, write one line describing what the reader of each channel is trying to do. GOV.UK's content design guidance on identifying user needs treats the job to be done as the starting point for every page, and the same order holds when you cut one article into several pieces. A LinkedIn reader scans for an opinion. A newsletter subscriber wants a link worth clicking.

Choose source articles that still work in six months

Repurposing rewards evergreen material. Look for posts with a clear argument, no dated news hook and no need for fresh interviews. A piece tied to a spring Budget announcement reads as stale by autumn, whatever the AI does with it. Keep a shortlist of about ten evergreen articles and work through it on a quarterly rotation.

Run the five-step repurposing workflow

Keep the steps in this order. Skipping the extraction stage is what makes AI drafts drift.

1. Pick. Choose an evergreen article with a clear argument, useful detail and no need for fresh interviews or updates. Note the audience for each planned channel and the reader’s job before you start; leave out anything that is stale or no longer supported.

2. Extract. Build a source sheet before drafting: capture the central point, supported claims and their evidence, examples, caveats, date-sensitive details and call to action. ChatGPT or Claude can help organise the source text, but ask for supporting passages and flagged gaps, then check the sheet against the article yourself. Do not paste confidential customer information unless your organisation’s approved account and data-handling terms allow it.

Prompt examples for step 3: In ChatGPT or Claude, try: “Using only the source text below, list its main point, factual claims with supporting sentences, examples, caveats, date-sensitive details and call to action. Flag anything the source does not establish; do not fill gaps.” For a draft, use: “Using only this approved source sheet, write [asset] for [channel] aimed at [audience]. Keep the stated facts and caveats, use British English, and mark missing information instead of inventing it.”

4. Check. An editor compares every claim with the source sheet, verifies names, figures, quotations and dates, and removes unsupported additions. Check paid promotional copy against the CAP Code, which the ASA enforces; for newsletter use, check the ICO’s UK GDPR guidance and verify that consent covers the messages sent. Approve the facts, tone and British English before anything goes live.

5. Schedule. Put each approved variant into your channel scheduler with its channel, publish date, owner and source article. Schedule only after sign-off, and keep the published copy beside the source so you can review its performance using the measures below.

Match each extract to the channel

Not every channel deserves the same version of the story. Match the extract to what that channel’s reader is trying to do:

For example, a small UK bicycle shop could turn an evergreen maintenance guide into a LinkedIn lesson and a newsletter care tip. A regional press pitch would need a genuine local angle; do not let AI invent local detail or data.

Regional titles rarely want a general version of a national story, so local detail does the heavy lifting. This is where AI content localisation pays off: swapping a city, a sector or a devolved policy reference can do more for a pitch than rewriting the whole piece. Email rewards a fixed structure, so keep the same three sections each issue and change only the content.

Set the guardrails before anything goes live

Copy that runs as paid promotion falls under the CAP Code. The advertising codes enforced by the ASA cover paid social posts as well as display ads. A discounted offer quoted in a LinkedIn ad needs evidence behind it before it runs.

Newsletters bring data protection into scope. If you are emailing a list, the UK GDPR guidance from the ICO sets out what you must tell people about how their details are used, and consent has to cover the messages you actually send.

Measure the pieces that earn attention

Impressions tell you very little at this scale. Track newsletter click-throughs, comments on LinkedIn and replies from journalists, and keep those numbers beside the source article so you can see which extract works. If one format wins twice, turn it into a template and retire the versions nobody reads.

Common questions

How much time should we budget per article?

Treat two hours as the working assumption for a two-person team on a first run. Later runs take less because the prompts, the style checks and the schedule are already in place.

Can AI turn a blog post into LinkedIn copy on its own?

It can produce a usable first draft, but the argument and the opening line still need a human. Posts that begin with a generic summary get scrolled past.

Do we have to tell readers that AI helped?

UK advertising rules focus on whether copy misleads, not on how it was drafted. If a piece carries a named author, that person should be able to stand behind every claim in it.

What if the source post was written with AI too?

Then the risk of repeated phrasing rises. Rewrite the variants from the extracted facts rather than from the draft, and read them aloud before publishing.