How we use AI
We use AI tools in some of our internal work. We disclose it. A human specialist reviews everything that reaches you and is accountable for it.
Where AI is used
In support of the specialist doing the work, never in place of them.
Internal drafting aids
An editor may use an AI tool to try alternative phrasings for a sentence, or to summarise a long source while preparing feedback. The wording that reaches you is written or rewritten by that editor.
Consistency and terminology checks
Tools help us catch a symbol defined twice, a unit that changes mid-paper, an abbreviation used before it is introduced, or a term that drifts between sections. Each flag is judged by a person against your field.
Formatting to author guidelines
Reference styling, heading levels, figure and table numbering, and layout against a specific journal or university template. Formatting is mechanical work, and it is checked by hand afterwards.
Every output is reviewed and edited by the specialist responsible for the job. Nothing is sent to you unread. If you ask what was used on your file, we will tell you.
Where AI is never used
These are refusals, not preferences. They do not move for a deadline.
- Generating substantive content that you would submit as your own
- Fabricating data, results, citations, or references
- Inventing sources, or filling a gap in your literature with text no one verified
- Any use a journal or institution would require you to disclose that we have not told you about
The argument you make, the data you report, and the final wording you submit remain yours. That is the line the rest of this policy protects.
Your disclosure obligations
Journals and universities increasingly require authors to declare AI assistance. The declaration is yours to make, because you are the author who signs the submission.
Language editing is usually treated differently from AI authorship. Most publishers ask you to acknowledge a professional editing service in the acknowledgements, and reserve the AI declaration for generated content. The requirement varies by publisher, by journal, and by year, so check the instructions for the venue you are submitting to rather than assuming.
We will tell you what we used on your job, in writing, so you can declare it accurately. If your journal wants a statement of our involvement, ask and we will put one together for that journal.
“The authors used a professional editing service (ThinkLab) for language editing and formatting to journal guidelines. The service used AI-assisted tools for grammar, consistency, and terminology checks. All changes were reviewed by a human editor and accepted or rejected by the authors, who take full responsibility for the content of the manuscript.”
This is an example to adapt, not a template to copy. Edit it so it describes what was actually done on your manuscript, and delete anything that did not happen. A disclosure that hides work is worse than no disclosure.
Detection tools
AI-detection tools are unreliable. They produce false positives on human writing, and they do so most often for writers whose first language is not English. Plain sentences, consistent phrasing, and careful academic register are exactly the features these tools score as machine-written. That is a flaw in the tools, not evidence about your work.
We do not sell “AI-detector clearance”, and we do not edit text to game a detector. Nobody credible can guarantee a detector score, because the tools are opaque, they change without notice, and their vendors do not stand behind the results.
A similarity report is a different thing. Where a service includes one, it is an iThenticate similarity report showing text overlap with existing sources. Similarity is not an AI-detection score, and neither is a certificate of anything.
If your institution questions your work, what helps is your record: drafts, notes, data, and version history. We keep tracked changes on the files we edit, so you can show what was changed and by whom.
Any service promising to make your writing pass an AI detector is promising an outcome it cannot control. We do not make that promise.
Confidentiality
Your manuscripts, data, and drafts are not used to train anything. They are not shared with other clients, sold, or published as samples. Access is limited to the specialist assigned to your job and the staff supporting it.
Files are destroyed on the retention schedule set out in the privacy policy. Unpublished results, patent-sensitive material, and student work under examination are treated as confidential by default, and we will sign a non-disclosure agreement if your institution requires one.