AI Content Editing

Which service do you need?

The two AI services do different amounts of work and the difference in price reflects it.

Why fact-checking is the core of it

Generative models produce fluent text regardless of whether the underlying claim is true, and they produce citations in the correct format regardless of whether the source exists. Both failure modes are invisible to a reader who is not checking.

In practice we find something in most drafts of any length: a statistic attributed to a report that does not contain it, a percentage that has drifted from the original, a quote attributed to the wrong person, or a study cited by a plausible-looking author, journal and year that cannot be found. Publishing any of those is a reputational problem that is entirely avoidable and very hard to walk back.

What the pass covers

• Every factual claim traced to a source, or flagged and removed.
• Every citation checked for existence and for saying what the text claims it says.
• Statistics checked against the original, including whether the figure has been rounded or reframed.
• Structural fixes: heading logic, section order, and removing paragraphs that restate the heading.
• Redundancy cut, which typically recovers 10 to 15% of the length.
• A source log returned alongside the copy, so you have the evidence trail if a claim is ever challenged.

What we actually find

These are the error types that recur, roughly in order of how often we see them.

• Statistics that have drifted. A real figure from a real study, quoted at a slightly different number or reframed to say something the study did not.
• Citations to sources that do not exist. Plausible author, plausible journal, plausible year, no such paper.
• Correct facts, wrong attribution. The claim is true and the person or organisation credited had nothing to do with it.
• Outdated regulation. Legal and compliance content confidently describing a rule that has since changed.
• Overstated consensus. "Studies show" or "experts agree" where the literature is contested.
• Fabricated quotes. Attributed, formatted correctly, and never said.

When editing is the wrong service

Sometimes a draft is not worth editing, and we will tell you rather than billing for a pass that cannot fix the problem.

The signals are: no usable structure, so every section needs reordering; a topic chosen without research, so the page targets nothing; or so many unverifiable claims that removing them leaves nothing behind. In those cases writing from a proper brief is faster and cheaper than editing, and we quote the writing rate with an explanation rather than the editing rate with a caveat.

What you receive

Three files, and the third is the one clients say they did not know they needed.

1. 1A tracked-changes version, so every edit is visible and reversible rather than presented as a finished fact.
2. 2A clean version ready to publish.
3. 3A source log: every factual claim in the piece, the source it was verified against, and anything removed with the reason. If a claim is challenged after publication you have the evidence trail rather than reconstructing it under pressure.

Turnaround and volume

Two working days for pieces up to about 3,000 words. Larger batches are scheduled rather than rushed, because fact-checking is the slow part and it is the part that matters.

For ongoing programmes we work to a standing arrangement: you send drafts as they are produced, we return them within two working days, and we invoice monthly on total word count. That works better than piece-by-piece ordering for teams publishing regularly.