How to Check AI-Generated Work Before You Send It to a Client
An AI draft can sound right and still be wrong. Check the brief, the claims, the sources, the numbers and the final file before it goes out.
The AI draft is clear, confident, and neatly arranged. That can make it feel more finished than it is.
Good writing is not proof that the facts are right or that the work answers the client’s request. Before sending AI-generated work, check the important claims against real sources, compare the output with the brief, and test the final deliverable.
You do not need to reject AI help. You need to know which parts you have verified and which still depend on a guess.
Begin with the brief, not the grammar
Read the request again. Who is the work for? What decision or action should it support? What must be included? What should be left out?
A polished guide for beginners may still be wrong for a client who asked for a technical comparison. A long report may fail because the client needed a short recommendation with clear reasons.
Write three or four checks before editing. For example: the proposal must reflect the agreed scope, use the current price, explain the delivery stages, and make no promise the team cannot meet.
Website refresh proposal
We will redesign the five agreed pages: Home, Services, About, Pricing and Contact.
The total cost is $4,800, as quoted in our March rate card.
Work runs in three stages: discovery, design, then build and launch.
We guarantee the new site will double your enquiries within a month.
These checks stop you spending an hour polishing the wrong answer.
Make a list of claims that need proof
Read the draft and mark statements a reader could reasonably ask you to support. Look for numbers, dates, product features, quotes, comparisons, and promises about results.
Separate them into facts you can verify, judgments you should explain, and assumptions you should label.
For example:
| Draft statement | What to do |
|---|---|
| “The product supports offline use” | Check current official product documentation |
| “Customers save ten hours each week” | Find relevant evidence or remove the claim |
| “This is the best choice for small agencies” | Explain the criteria and limits of the judgment |
| “The client will provide images by Tuesday” | Confirm it or label it as an assumption |
Open the source and check what it actually says
A citation can be real and still fail to support the sentence. Open it. Check the author or organization, date, scope, and relevant passage.
If the draft says a study proves a result for small businesses, make sure the study actually included the relevant group and measured that result. If it says a product has a feature, use the provider’s current documentation rather than a loose summary from another website.
“A 2024 study proves that weekly reviews raise revenue for small businesses.”[1]The citation is real. That doesn’t mean it says this.
When you cannot verify a claim, remove it, narrow it, or state the uncertainty. Do not ask AI to invent a better-looking citation.
For work involving legal, medical, financial, or other high-consequence claims, get the appropriate specialist review. A general writing check does not establish professional accuracy.
Recalculate numbers yourself
Check totals, percentages, units, and dates. A draft may use the right inputs but describe the wrong comparison.
The AI draft says this is “a 5% increase.” What is it really?
| Spreadsheet test | Input | Expected | Formula gave | |
|---|---|---|---|---|
| Normal case | 20 → 25 | 25% | – | – |
| Zero items | 0 → 5 | “New” | – | – |
| Blank field | (blank) → 5 | “No data” | – | – |
For a simple example, moving from 20 enquiries to 25 is an increase of five enquiries, or 25% relative to the original 20. It is not an increase of five percent.
For a spreadsheet, test formulas using inputs with an answer you already know. Check a normal case and a boundary case, such as zero items or a blank field. For code, use appropriate tests and review rather than relying on an explanation that sounds plausible.
Look for missing context and false certainty
AI can fill gaps with a reasonable-sounding story. Watch for details the client never supplied: audience size, budget, launch dates, customer needs, or product capabilities.
The client’s brief only said: “Plan a spring email campaign for our existing customers.” Click every detail the AI made up.
Change “The campaign will increase sales” to a supported, honest statement such as “The campaign is intended to increase qualified enquiries; we will compare results after launch.” Do not promise an outcome the evidence cannot guarantee.
Also check whether the draft presents one option as the only option. A useful recommendation usually explains what would make a different choice better.
A small before-and-after example
Consider this fictional draft line. The team has not tested the guide. Drag the handle to see a more honest version.
The second version is still useful. It names what was done and how the team will judge it, without pretending a result has already happened.
Check the final file, not just the draft
Open the exported document, image, slide deck, or published preview. Check links, layout, readability, and access. Read it once as the intended recipient.
Ask: What would I do next? What would I still need to ask? Is any private note or placeholder still visible?
Website refresh proposal
Thank you for the conversation last week. This proposal covers the five agreed pages and three delivery stages.
Total: $5,200 for discovery, design, and build.
To accept, sign here:
Use AI review as another check, not independent proof
M-Tracker’s work quality review can help compare completed work with the task brief and point out apparent gaps. Keep the brief and readable evidence together, then inspect the feedback before deciding what to change.
Another AI response does not independently verify the first AI response. If both repeat the same unsupported claim, you still need to check the source. Use M-Tracker for fit and review support; retain human responsibility for facts and final delivery.
Get started for freeNo credit card required. No employee surveillance.
A final handoff checklist
Before sending, be able to answer yes to these questions:
Common questions
Do I need to rewrite everything AI produced?
No. Keep useful work that meets the brief and passes your checks. Editing for its own sake is not the goal.
Is a plagiarism or AI-detection score enough?
No. A score does not establish factual accuracy, permission to use material, or fit with the brief. Review the work itself and any source material it relies on.
What if I cannot verify a key claim before the deadline?
Tell the reviewer what remains uncertain. Remove the claim or delay that part of the work rather than presenting it as settled fact.
Work you can trust, without redoing it.
Your team gets feedback before work reaches you, or your client.
Copyright © 2020 – 2026 M-Tracker. All Rights Reserved. Privacy Policy