Writing review guides · Draft comparison
Did an AI Rewrite Change Your Meaning? A Draft Comparison Checklist
To check whether an AI rewrite changed your meaning, compare the claims as well as the wording. Mark what became stronger, broader, more precise or differently attributed. Keep a correction only when it expresses what you intend and the available evidence supports it.
Published 15 October 2026 · Publisher: Ethara Global Technologies Inc., developer of VEYTHARA by ETHARA™.
Method: the original paragraph, supplied source and revisions below are fictional editorial examples created for this guide. They are not observed AI outputs or results from a real study.
Seven changes a text diff can miss
A diff highlights additions and deletions. It does not establish their significance. Read each changed sentence with these seven questions in view; a single removed word can reverse a claim.
| Check | Example change | Question to resolve |
|---|---|---|
| Negation | “did not improve” → “improved” | Did a negative become a positive, or an exception disappear? |
| Certainty | “may help” → “will help” | Does the evidence justify the stronger confidence? |
| Quantities | “up to 12%” → “12%” | Did an upper limit become an exact result? |
| Dates | “in the March pilot” → “currently” | Has a limited observation become a present claim? |
| Causation | “fell during the trial” → “the tool reduced” | Was a cause established, or only a change observed? |
| Population and scope | “one support desk” → “all teams” | Does the revision extend beyond the people or conditions studied? |
| Attribution | “staff reported” → “we confirmed” | Who observed, claimed or concluded this? |
Compare an original paragraph with two revisions
Supplied source — fictional pilot note, paragraph 2: “In March, one support desk tested a new checklist. Staff reported a maximum 12% shorter handling time on selected routine tickets. There was no comparison group. Customer satisfaction was not measured.”
Original draft: “The March pilot results suggests that the checklist may shorten handling time by up to 12% for selected routine tickets at one support desk, according to staff reports. It did not measure customer satisfaction.”
Grammar-only revision: “The March pilot results suggest that the checklist may shorten handling time by up to 12% for selected routine tickets at one support desk, according to staff reports. It did not measure customer satisfaction.”
Overconfident revision: “The checklist reduces handling time by 12% across support teams and improves customer satisfaction.”
The grammar-only version corrects agreement. The overconfident version removes the date, upper limit, selected-ticket restriction and staff attribution. It also adds a satisfaction claim that the source does not address. Fluency alone cannot resolve those differences.
Build a claim-by-claim decision table
Record the original span, revised span, change category and decision. “Accept” means you checked that change, rather than every nearby sentence.
| Original → revised | Change and reason | Reviewer decision |
|---|---|---|
| “results suggests” → “results suggest” | Agreement correction; qualifications remain. | Accept the verb correction. |
| “may shorten … up to 12%” → “reduces … by 12%” | Certainty, quantity and causation become stronger. No comparison group is supplied. | Rewrite: “Staff reported up to 12% shorter handling time during the March pilot.” |
| “selected … at one support desk” → “across support teams” | The population is broadened without evidence. | Reject the broader scope; restore the restriction. |
| “did not measure” → “improves customer satisfaction” | An unmeasured outcome becomes a positive result. | Reject and remove the added outcome. |
For your own worksheet, copy these fields: draft/version · original span · revised span · category · source/location · accept, reject or rewrite · reason · reviewer/date. Keep uncertain points open rather than forcing a yes/no answer.
Bring sources back into the comparison
Find the passage for each factual assertion. A passage that does not discuss satisfaction is missing evidence for an improvement claim; it does not show that satisfaction worsened. By contrast, a sentence saying satisfaction was not measured directly conflicts with a rewrite saying the pilot measured it.
Read the surrounding paragraph and preserve limitations. Use the claim–evidence worksheet to separate support, contradiction and unanswered questions.
What the software flags and what you decide
VEYTHARA’s draft-comparison workspace displays wording differences and literal correspondence with supplied source passages. Rule-based prompts point to places to inspect. It does not authenticate sources or automatically determine semantic equivalence. A repeated phrase may be relevant without supporting the entire claim.
Review the highlighted text, inspect the source and decide whether to retain, reject or rewrite it. A saved case keeps supplied drafts, versions, owner notes and decisions. That record documents the available material and your choices; it does not certify correctness, authorship or a complete writing history.
Keep a record of the final choice
A defensible final example is: “During a March pilot at one support desk, staff reported up to 12% shorter handling time on selected routine tickets. The pilot had no comparison group and did not measure customer satisfaction.” The reviewer accepts the grammatical correction, narrows the outcome wording and records why the added satisfaction claim was removed.
Keep the original, revision, source identifier and note together. Check the final paragraph again: small changes can combine into a different message.
For a new revision, try the free writing test. For grammar choices, read how to fix grammar while keeping your voice.
A text-and-evidence review pilot by Ethara Global Technologies Inc.
Final judgments about evidence, meaning and attribution remain with the reviewer.
