AI tools write generic social copy when the only guidance is “friendly but professional”, because that describes almost every business. To keep drafts sounding like you, write a one-page voice sheet with rules a reviewer can check and real posts as examples, give it to the tool every time, and add a rule whenever you correct the same thing twice. Check claims separately: a post can sound exactly right and still say something you cannot back up.
This applies to any post, not just ones made from an existing article. If you are turning a blog post into a set of posts, our guide to repurposing an article with AI covers the extraction step. This guide is about what keeps every post consistent once drafting starts.
Turn adjectives into rules someone can check
| Vague description | Rule a reviewer can check |
|---|---|
| Friendly | Write to one reader as “you”. Use staff first names. No exclamation marks in the first line. |
| Expert | Each post explains one specific step, fact or example. Never call ourselves “leading” or “world-class”. |
| Local | Name the town or area when it matters to the reader. Do not claim to cover places we do not serve. |
| Straightforward | Sentences under 20 words. One idea and one call to action per post. Prices include “+ VAT” where it applies. |
| Not salesy | No countdowns or “don’t miss out”. No more than one question in a post. |
If you have a brand brief, start from its personality section. Our brand brief template asks for three words you are and three you are not; the “not” list often turns straight into banned words and phrases.
Build a voice sheet from posts that already work
- Collect 10 to 15 approved posts Posts you would happily publish again, across your main channels. Real examples beat descriptions.
- Write down what they have in common Sentence length, how you open, how you refer to customers, how you end. Describe what the posts actually do, not what you wish they did.
- List words to use and words to avoid Include the phrases AI tools reach for that you never use, such as “unlock”, “elevate” or “game-changer”, and your own preferred terms (“customers”, not “clients”, or the reverse).
- Note how tone shifts by situation A launch, a practical tip and an event reminder can differ in energy without changing vocabulary. Replies to complaints should be written by a person, not drafted from the sheet.
- Add channel notes, not a different voice LinkedIn might allow a longer explanation; Instagram captions might be shorter. The voice stays the same.
- Include one or two off-brand examples Label exactly what is wrong with each. Contrast teaches a tool more than another adjective.
- Give it an owner and a date One person updates the sheet. Version it, so you know which rules a draft was written against.
Keep it to a page. A twelve-page guideline will not be read by your team, and long instructions are followed less reliably by AI tools too.
Check claims separately from voice
Voice review asks “does this sound like us?” Claim review asks “can we prove this?” They need different checks, and the second matters more. Prices, numbers, comparisons such as “cheapest” or “fastest”, guarantees, results, availability and testimonials are the claims to watch.
In the UK, the CAP Code applies to marketing on a business’s own website and in other online space it controls, including its social media channels, where the content is directly connected to selling its goods or services. Before publishing, you must hold documentary evidence for objective claims, and evidence that testimonials are genuine. The ASA’s guidance on its remit for own websites and social media explains where the line sits. This is general information, not legal advice.
The practical answer is an approved-claims list kept alongside the voice sheet: each claim, where it comes from, and when it needs rechecking. Tell the tool to use only those claims and to mark anything else as needing a check. Never let it invent a statistic, a customer quote or a named person.
A worked example
A fictional physiotherapy clinic’s voice sheet says: plain, calm, specific; no exclamation marks; no superlatives. Its approved claims are: first appointments are a 45-minute assessment; evening appointments until 7pm on weekdays; online booking.
First AI draft: “We’re the leading physio clinic in the area! Book now and feel better in days!!” It breaks three voice rules (superlative, exclamation marks, hype) and makes two claims that are not on the approved list: “leading” and “feel better in days”, which is an outcome promise.
Edited post: “Back pain making desk days hard? Your first appointment is a 45-minute assessment, so treatment is based on what’s actually going on. Weekday appointments until 7pm. Book online.” Every fact comes from the approved list, and the tone matches the sheet. The reviewer then adds “no outcome timescales” to the banned list, because it is the second time it has come up.
Make the review quick and the approval real
- Does it sound like the example posts? Read it aloud.
- Any words from the avoid list?
- Is every price, number, comparison and testimonial on the approved-claims list, and still in date?
- No invented quotes, reviews, people or results?
- Does the call to action go to a page that works?
- Does it fit the channel and the template it will sit in?
- Who approved this exact version? If the text changes after approval, it needs approving again.
Give one named person the final say for each post. A tool can check for banned words or flag a claim that is not on the list, but it cannot take responsibility for what you publish.
Keep the look as consistent as the words
Consistent copy on inconsistent graphics still looks like five different businesses. Reusable templates for your most common post types, with fonts, colours and photo style written on one page, do for the visuals what a voice sheet does for the words. That is what Foundry’s social media graphics work provides: designed templates your team can edit, plus a short guide. We design; you or your team stay in charge of what gets posted.
