Thinking of using AI to Write for DFTB?

Team DFTB

doi:10.31440/DFTB.68686

Home > Thinking of using AI to Write for DFTB?

Thinking of using AI to Write for DFTB?

Team DFTB

Written on 02/07/2026
, Last updated 02/07/2026

Cite this article as:

Team DFTB . Thinking of using AI to Write for DFTB?, Don't Forget the Bubbles, 2026. Available at: https://doi.org/10.31440/DFTB.68686

SHARE VIA:

Our ethical framework

A reader trusts a DFTB post the same way they’d trust a colleague at handover. They assume a real clinician stood behind every sentence, checked every dose, and read the paper before summarising it. AI tools can now write something that sounds exactly like that colleague, whether or not one was involved. That gap between sounding trustworthy and being trustworthy is where an ethics policy earns its keep.

We didn’t want to invent our AI policy from scratch. Peer-reviewed medical journals have spent the last two years arguing this out in public, under far more scrutiny than a blog ever faces. So we read the author guidelines from the journals a paediatric clinician actually cites: the ICMJE recommendations that most medical journals follow, Pediatrics and the AAP journal family, JAMA Pediatrics, The Journal of Pediatrics (Elsevier), the Journal of Paediatrics and Child Health (Wiley), Pediatric Research (Springer Nature), NEJM, and Archives of Disease in Childhood and BMJ Paediatrics Open (BMJ). Despite different publishers and different lawyers, they land on almost the same rules. That is the basis for what follows.

What the journals agree on

Strip away the house-style differences, and every one of these guidelines says the same five things.

AI doesn’t get a byline. Authorship means taking responsibility for accuracy and integrity, and a model can’t do either. Every journal we checked, from ICMJE down to NEJM, states this outright: AI tools cannot be listed as authors or co-authors.

Disclosure is not optional. If AI helped write, structure, or edit a piece, journals require authors to say so, usually in a dedicated statement. Elsevier’s language is the clearest: authors must document “the name of the AI Tool used, the purpose of the use, and the extent of their oversight.” Grammar and spell-check tools are typically exempt. Drafting help is not.

The human checks every fact. This is the line every journal returns to. AI-generated content can sound authoritative while being wrong, and AI-generated citations can be fabricated. Authors, not the tool, are accountable for verifying accuracy before anything is published.

No AI-generated clinical images or data. Journals draw a hard line here: AI tools can help with prose, but not with creating or altering figures, images, or data that represent real findings.

Privacy comes first. Several policies explicitly warn against pasting identifiable patient information into an AI tool. Once it’s in, you’ve lost control of where it goes.

Our principles at DFTB

We write for clinicians making decisions about real children. We set a higher bar than most blogs, so we’re adopting the journal standard rather than a lighter one.

No AI bylines, ever. A DFTB post is written by the named author. AI can be a drafting tool, never a contributor.
We say when we’ve used it. If AI has done more than tidy a sentence, that’s a paragraph, not a footnote, and it goes in a short disclosure at the end of the post.
Every claim, dose, and citation gets checked by a human who read the source. Not skimmed. Not “the AI summarised it for me.” Read. This is non-negotiable for anything clinical.
AI can help with structure and clarity, not clinical judgement. Use it to find a better opening line, tighten a paragraph, or suggest a rebuttal to a difficult argument. Don’t use it to decide what the evidence means.
Patient and family information never goes into an AI tool. No case details, no identifiable quotes, no exceptions.
The voice stays human. DFTB has 580-plus authors and no house voice, only a house floor: plain language, active voice, no padding. AI drafts tend to flatten that. Edit until the post sounds like you again.

What this looks like in practice

Before you submit a post that used AI for more than copyediting, add a short disclosure. Something like:

AI assistance disclosure: [Tool name] was used to [draft an initial outline / suggest structural edits / tighten prose] for this post. All clinical content, references, and conclusions were written and verified by the author.

And run this check before you hit publish:

Have I read every source I’ve cited, not just trusted a summary of it?
Would I be comfortable if a reader asked me, “Did you write this”?
Does this still sound like me, not like a press release?
Is there anything here a family or a colleague could act on if it were wrong?

If the answer to that last one is yes, the bar for checking it is not “good enough,” it’s “correct.”

Frequently asked questions

Does DFTB use AI to write its blog posts? Sometimes, for drafting help, structure, and editing. AI never decides the clinical content, and it’s never listed as an author. Every post is written and owned by a named clinician.

How does DFTB disclose AI use in a post? Any post where AI did more than copyediting carries a short disclosure statement naming the tool, what it was used for, and confirming a human verified the clinical content.

Can AI write medical content safely? Only with a human checking every fact, dose, and citation against the original source. AI-generated text can sound authoritative while being wrong, and AI-generated references can be fabricated, so unchecked AI content is not safe to publish in a medical context.

What do medical journals say about AI-generated content? Every major journal we reviewed, including ICMJE, JAMA, NEJM, and BMJ, agrees on the same core rules: no AI authorship, mandatory disclosure, and full human accountability for accuracy.