AI agent for reviewing content before it goes out
This is the agent of the last step, not the first. The text is written, agreed and all but sent — and this is the cheapest possible moment to notice that general information has turned into professional advice, that nothing backs the number in the second line, and that the paragraph about your approach is nearly word for word what went out for a different client last month.
What the agent checks before anything is published
- Reads the finished draft end to end and marks what it finds, instead of rewriting it for the author.
- Catches wording that has crossed from general information into professional advice.
- Tests every claim against what actually backs it — a figure, a timeframe, a «first in the city» — and flags the ones nothing supports.
- Checks the disclosures your trade expects are present, and names the one that is missing.
- Compares the text with what you have already published and highlights the overlap, including a different client's.
Processes that run themselves
A playbook isn't a button — it's a process with a condition that starts it. These three are the reason the review exists as a separate step at all.
A draft reaches the review step
A text is finished and handed to the last step before publication
- Read it whole and mark the passages that read as advice rather than information
- Match every claim against what backs it, and flag the ones that stand on nothing
- Check that the disclosures your trade expects are in place
- Send it back to the author marked up, without changing a word of it
A case study is prepared for publication
A client's result is being written up as a public case
- Check that consent to publish exists, and what exactly it covers
- Flag promises of a guaranteed outcome and comparisons with named competitors
- Reconcile the numbers in the story with the numbers in the reports, and list the gaps
- With no consent, stop — and do not offer a workaround
The same recommendation appears for a second client
A draft proposes an approach you have already sold to another client
- Compare it with what has already gone out under other clients' names
- Show the matching passages side by side, not a similarity score
- Leave you the decision on whether it changes, and what changes
Where the agent stops
- The agent marks and reports. It clears nothing for publication: a person presses the button, and that is the whole point of the arrangement rather than a formality.
- This is not legal advice. The agent does not replace a lawyer, a doctor or a compliance officer — it prepares the text for their review, it doesn't stand in for it.
- Where the rules turn on your trade or your country, that judgement stays with you and your adviser. The agent describes what it found; it doesn't rule on it.
- On similar input it produces similar output. That makes it a steady check rather than an opinion: it will never tell you the text is good.
- Responsibility for what gets published stays with the business. Having a review step does not move it or share it.
FAQ
Will it edit my text itself?
No, and that is deliberate. It hands the draft back marked up: advice instead of information here, a figure with no source there, this paragraph has already gone out. An agent that edited and published in one pass would leave the text with no reader again — and the whole value of the role is that somebody read it before it went out.
Does this replace a lawyer or a compliance review?
It doesn't. The agent takes the mechanical part off the table — the missing disclosure, the promise of a result, the claim with nothing behind it. What is left is the judgement, and that stays with the person who is professionally answerable for it. It just reaches them already worked through instead of cold.
How does it know a text repeats something already published?
It compares the draft against what you've published before — your own material and other clients' material it has access to. It shows the matching passages side by side rather than a percentage, because similarity on its own isn't the problem. The problem is a client recognising a competitor's strategy in their own.
What if it flags something that was actually fine?
It will, and the bias is intentional. A flag costs you thirty seconds of reading; a missed line costs a conversation with a client or worse. A flag doesn't block publication — you look and move on. What it should stop flagging gets tuned as you go.
Where this shows up
Walkthroughs where the review sits as the last step before anything goes out.
Running a dental practice's social media when the dentists won't spend time on it
You have cases the practice down the road doesn't. What's missing isn't material — it's someone to build posts out of it without writing clinical nonsense.
Read the walkthrough →Agencies & servicesAn agency run on AI agents: where strategy has to go back to a human
Agents carry production well. The problem is elsewhere: strategy starts looking the same across every client, and clients notice before you do.
Read the walkthrough →