Agency Workflows & Quality
How to spot AI writing: read the process page.
You cannot settle the question by reading the prose. Read the page where the publisher says what a tool did and what a person checked, then run the same five checks on your own draft.
You do not trust a file you cannot trace
You have opened a free workbook, read two pages, and felt the floor tilt: headings in a neat row, a paragraph that restates the title, and no sign that a person with clients ever touched it. You closed the tab and did not send it to anyone.
The fear underneath is about your name, not machines. If you adapt a file like that and a senior reader finds a page saying nothing, the damage lands on you.
So the useful question is not whether a page sounds generated. It is whether you can see how it was made. A publisher who shows the process gives you something to check; one who shows only a finished document is asking for trust you have already spent.
What follows is a five-item read of any publisher's process page, then the same checks on your own draft before it goes out under your name.

Why AI content detection settles so little
Read two pages and you will find signals pointing both ways. Flat sentence lengths and three-item lists appear in machine drafts and in human compliance writing. A specific, lively paragraph can come from a consultant with twenty years of notes or from a careful editor working on a machine draft.
AI content detection tools narrow this slightly and settle nothing. They return a likelihood score built from patterns in the text. A score is not a source, two tools can disagree about the same paragraph, and a confident number says nothing about whether the advice is sound.
Platform labels carry the same limit. Meta's “Made with AI” labels on video, audio and images are based on its detection of industry-shared signals of AI content or on people self-disclosing that they uploaded AI-generated material, and Meta says it may add a more prominent label where content creates a particularly high risk of materially deceiving the public. That is a statement about origin and context. It is not a check on a single sentence of advice.
That leaves the process page: the one place where a publisher can say, in plain words, which steps a person owned. It is short, unglamorous, and the most useful page on most sites.
Build your brief as you read.
Use the free 25-page fillable Lead Magnet Brief Builder: 19 lettered worksheets from finding an idea to a one-page brief, a draft skeleton and a launch plan, with three fictional worked examples.
Free: we send you a short email first. Reply VERIFY to it and we email you the workbook, then five short follow-up emails over 12 days (one names our own $79 Workflow Pack), then at most one email a week, when we publish a new guide. Every email has an unsubscribe link.
The five-item read of a process page
An editorial or process page does not need length. Five questions, each answerable in a line, tell you whether the publisher has thought about this at all. Open their page in a second tab and work down the list.
1. Who publishes it, and where do I write to?
Look for a publisher named in plain words and a contact route you could use: a form, or a page saying who answers. You are not hunting for an impressive legal entity. You are checking that someone is on the other side and can be asked a question. No publisher and a contact form that vanishes is an anonymous document with a logo.
2. What did the AI do, and what did a person do?
The page should split the work: which steps a tool drafted, which a person wrote or rewrote, and where the judgement sits. Two sentences are enough. Illustrative page line: “Drafts are assembled with AI assistance from a written brief; the examples, the decision rules and the final edit are the publisher's.” A page that says “AI-powered” and stops has told you nothing you can check.
3. Who checked the claims, and what did the check look like?
You want the check, not a job title. “Every figure is compared with the cited page on the date shown” tells you what happened. “Reviewed by experts” tells you who was in the room and nothing about what they did. Where a reviewer is named, look for which claims they checked and whether they could remove a sentence.
4. Which sources are cited, and are they dated?
A citation you can open is the difference between a claim and an assertion. Look for the link, a retrieval date, and a line saying what the source was used for. Undated links rot quietly: a price page from two years ago still loads and still looks authoritative. Our guide to citing sources in a lead magnet sets out the method inside the file itself.
5. How do I ask for a correction?
This is the item most publishers skip. You are looking for a stated route and a plain sentence that errors will be fixed, not a response time nobody can honestly guarantee. A page with no correction loop keeps every mistake it has ever had.
Only one of the five is about technology. The other four are about accountability, sourcing and repair, and they are the ones that protect your name when you adapt someone else's file.
What that looks like written down
Our own Editorial Policy states who publishes this site, that pages are drafted with AI assistance and checked against the policy and the cited sources, and that the contact form is the route for questions and corrections. The Review Methodology covers the other half: tool pages summarise vendor documentation, and no tool has been tested for this site.
You can audit that in two minutes. Open a tool page, find a sentence about a vendor feature, follow its source link and retrieval date, and compare the two. If the source does not support the sentence, you have found something real, and the correction route is right beside it.
Run the same five checks on your own draft
Before you adapt anyone's workbook, write one paragraph for your own site. Four lines, no software:
- Where the material came from, and that you have permission to use it.
- Which parts a tool drafted and which parts you wrote or rewrote.
- How each figure and comparison was checked, and against what.
- Where a reader sends a correction.
Then run the claim pass. Read the draft once asking only which sentences state a fact, a figure, a comparison or an outcome, and sort each into four piles.
Keep
You can point to the source in one click.
Verify
A document or a check is needed first.
Revise
The point holds; the wording overclaims.
Remove
Nothing supports it and nothing needs it.
Watch the soft claims, which arrive without a citation and sound modest: “most teams”, “studies show”, “in today's market”. A draft produces them when no source was supplied. Rewrite them as your own practice, or cut them.
“In today's fast-moving business environment, robust onboarding processes are essential for seamless alignment.” Rewrite: “Ask your last three new managers which week felt hardest, and write down what they say.”
The Source & Claim Review Checklist (version 2.1, October 2026) turns this into ten checks and three claim records you keep beside the draft: one row per claim, the source, the date and the decision. For the longer argument, see reviewing AI-written claims before you publish.

Will a checklist like this just collect dust?
Sometimes, yes. A general checklist stays generic until you put your own facts into it, and that is as true of ours as of anyone's. Three edits do the work: replace one example with a situation from your own client work, with no names or identifying detail; add the decision rule your niche uses where the checklist gives none; and cut two items that do not apply, adding one that does.
Illustrative edits. For an HR consultant, “confirm the process owner” becomes “confirm whether the line manager or HR signs the revised job description”. For an operations consultant, “check the data” becomes “check whether the supplier quote includes the switching cost”. Neither needs a tool, and both make the file yours.
Your next step: open one process page today
Take the last free workbook you downloaded and find the publisher's editorial or about page. Score it against the five items. If it fails two, do not adapt it for a client. If it passes four, keep it, and copy the shape of its process note for your own resource.
Then write your four lines. Tool-drafted or hand-written, the note is the same length either way. It is the difference between a file with your logo on it and a file you can answer questions about.
Free: the Source & Claim Review Checklist
A two-page fillable PDF with ten review checks, a worked example and three claim records. Pair it with our Editorial Policy, which sets out who publishes this site, how pages are drafted with AI assistance, and where a correction goes.
Free · email to download: enter your email and reply VERIFY to the short email we send. We email you the checklist, then five short follow-up emails over 12 days, then at most one email a week, when we publish a new guide. Every email has an unsubscribe link.
Quick questions
How can you spot AI writing without a detector?
You cannot prove authorship from the prose. Read the publisher's process page instead: who publishes, which steps the AI did, who checked the claims, which sources are cited with retrieval dates, and how to request a correction. Those five answers tell you more than any score.
Are AI content detection tools reliable enough to decide?
No. They return a likelihood score built from patterns in the text, and two tools can disagree about the same paragraph. Treat a score as one weak signal; if it decides whether you publish or adapt a document, you are guessing.
Where should my process note live, and what goes in it?
On the editorial or about page, linked from the footer, in one paragraph: source and permission, what a tool drafted, how claims were checked, and where corrections go. Update it when the process changes.
Does a platform label settle whether content is trustworthy?
No. Labels such as Meta's ‘Made with AI’ are based on detecting industry-shared signals or on self-disclosure, and they speak to origin and context, not to whether a claim inside the document is accurate.
Sources & editorial notes
- Meta: our approach to labelling AI-generated content, retrieved 6 November 2026. Used only for these points: “Made with AI” labels on video, audio and images are based on detection of industry-shared signals of AI content or on self-disclosure, and a more prominent label may be added where content creates a particularly high risk of materially deceiving the public.
- Resource contents described from this site: Source & Claim Review Checklist, Editorial Policy and Review Methodology.
- No tool was tested for this article. The page lines, the four-line process note, the rewrite and the two edits are illustrative, not client results. This page contains no affiliate links.
Examples are illustrative. See our Editorial Policy, Review Methodology and Affiliate Disclosure.
Read the process page, then test one claim.
The free Source & Claim Review Checklist gives you ten checks and three claim records to keep with a draft. The Editorial Policy shows how this site answers the same five questions.
Free: we send you a short email first. Reply VERIFY to it and we email you the workbook, then five short follow-up emails over 12 days (one names our own $79 Workflow Pack), then at most one email a week, when we publish a new guide. Every email has an unsubscribe link.
