A chatbot, an absent byline and a citation request
You are a senior medical writer at an agency. A client's medical affairs team has commissioned a review article on real-world evidence for a biologic in severe asthma, contracted to a journal that follows ICMJE recommendations. Two weeks before the deadline the lead writer admits she drafted three sections by pasting unpublished, patient-level pharmacokinetic data and a table from an internal clinical study report into a free public generative AI chatbot to clean up the language. The client's medical director now says three things: the writer's name should be left off the byline because agency writers are not real authors; a key opinion leader should be listed as first author although he reviewed the draft once; and the KOL will only agree to be involved if the manuscript cites four of his papers, two of which are only tangentially relevant. The account director wants the paper submitted on time and asks you to handle the details.
1. How do you handle the AI tool use? Address confidentiality, disclosure and what has to be redone.
Read the worked answer
Treat it as a data incident first. Unpublished patient-level PK data and an internal CSR table were entered into a public tool. That is a potential breach of the confidentiality agreement, of data protection obligations for identifiable or re-identifiable patient data, and of the client's own data-handling policy. Notify the client's data protection or privacy contact and your agency's compliance lead the same day; do not wait for the article to be finished. Preserve what was entered and when, so the exposure can be assessed.
Then deal with the content. Text derived from that input cannot be assumed to be accurate - chatbots can drop, merge or fabricate detail, including references. The three sections must be rewritten by a human from the source documents. Every reference in the rest of the draft needs verification against the actual article, not the chatbot's rendering, because hallucinated citations are the most common failure mode.
Disclosure. ICMJE guidance and GPP2022 both require transparency about AI assistance. If any AI tool is used, it is disclosed in the methods or acknowledgements, describing the tool and what it was used for; AI cannot be an author, and authors remain fully accountable for all content. A tool used to 'clean up language' still counts and still must be declared, unless the journal's policy says otherwise.
Prevention. Ask the client and agency to agree an approved-tools list, a rule that no unpublished or patient-level data goes into any external tool, and a short attestation at submission.
Timeline honesty. Tell the account director the two-week deadline may slip. A retraction or a privacy complaint costs far more than a delayed submission.
2. The client wants the writer's name off and the KOL listed as first author. How do you resolve authorship?
Read the worked answer
Apply the criteria, not the job title. ICMJE sets four: substantial contribution to conception or design, or acquisition, analysis or interpretation of data; drafting or critically revising for important intellectual content; final approval of the version to be published; and agreement to be accountable for all aspects. All four must be met. GPP2022 is consistent with this and explicitly rejects ghost and gift authorship.
The writer. A professional writer who drafted and revised the review, verified the data and will approve and stand behind the final version substantially meets the criteria. Deliberately omitting a qualifying contributor is ghost authorship, and 'agency writers are not real authors' is not a justification - it is the classic form of it. Discuss with the client: either the writer is listed (position determined by contribution, not seniority), or the writer's role does not meet the criteria and is disclosed in the acknowledgements with funding. What is not acceptable is a qualified contributor left invisible while someone else takes the credit.
The KOL. A single review of a draft does not satisfy criterion 1 or criterion 4. He cannot be first author. If he contributes substantively - shapes the question, interprets the evidence, revises critically, agrees to be accountable - authorship can follow. Otherwise he is acknowledged.
The institutional issue. The client is asking you to sign off on a byline that misrepresents contribution. Say no in writing, with the criteria quoted, and copy the agency's compliance contact. Also check what the journal's authorship declaration form will require; a false declaration is a research-integrity matter, not a negotiation.
3. How do you respond to the KOL's demand that four of his papers be cited, two of them only tangentially relevant?
Read the worked answer
Separate legitimate from coerced citation. Citing an author's work because it is directly relevant to the review's question is normal and expected. Citing it to secure participation is not, and it is a recognised form of citation manipulation - it distorts the evidence base and, if it came to light, would damage the KOL, the client and the journal.
Practical handling.
- Assess each of the four papers on merit. Include any that genuinely bear on the review question, with a sentence on what each contributes.
- For the tangential two, explain in writing why they are not included - a review that cites everything relevant to the author rather than to the question is a weaker review.
- Offer alternatives that are not coercive: an acknowledgement, a review of a section, a quoted perspective in the discussion, or co-authorship if he meets the criteria.
- Make clear the final reference list is the authors' responsibility and cannot be a condition of authorship or participation.
If he withdraws. Accept it. A review can be completed without him, and the client should be told that the alternative is a compromised manuscript. Document the exchange so the decision is attributable.
Wider control. Add a line to the author agreement: authorship and participation are not contingent on citation of specific works, and the reference list reflects the evidence, not the negotiation. Also check the client's own standard operating procedure, because this request often originates with a field team rather than the KOL himself.