What happened
On August 17, 2026, The Wall Street Journal reported that generative AI is forcing the book industry into a practical authorship crisis. The story described big publishing deals that fell apart after suspected AI use made agents or publishers uncomfortable with standing behind the work.
One example in the report involved Jerry Falade's debut crime novel, Call Me, I'll Hide the Body. WSJ reported that after a multimillion-dollar deal, the author's agents withdrew because they could not verify that the manuscript had been wholly written by Falade. The report also connected the case to earlier controversy around Mia Ballard's Shy Girl.
The Authors Guild's primary material shows why the issue has moved from social suspicion into publishing process. Its Human Authored certification lets authors certify that a book's text was written by a person, with only limited exceptions such as spelling, grammar, or research assistance.
The Guild's April 2026 model clauses go further. They address publisher use of AI, author use of AI, AI training, RAG, AI summaries, chat-with-a-book products, translations, audiobooks, and substantive manuscript editing. One clause says authors should disclose any AI-generated text and keep it below a minimal amount if the work is being represented as original to the author.
For expert publishers, the authorship question is what the named person contributed. Record that contribution alongside any AI assistance. Book-contract requirements need separate review.
Document who supplied the claim
Keep enough evidence to show who supplied each claim and who reviewed it.
AI can help a team prepare questions, organize research, clean transcripts, draft outlines, suggest titles, and turn an approved answer into different formats. Those uses can save time without pretending the tool lived the experience, made the judgment, or owns the claim.
The trust problem starts when the final asset implies a human point of view that the human did not actually supply. In books, that can become a dispute about whether the manuscript is original to the author. In expert content, it can become a quieter problem: the post sounds like a founder, researcher, consultant, or creator, but nobody can point to where they actually said the thing.
If a byline, face, or voice implies that a person believes, tested, learned, or recommends something, trace that claim back to the person and their sources.
Disclosure still matters. But disclosure tells people a tool was involved. Source proof tells people what the person contributed and what the team checked.
Why interviews help
A research-guided video interview creates a source before the article, clip, caption, or newsletter exists. The team starts with context and public sources, prepares better questions, and records the person answering in their own words.
That recording becomes the source of record. The transcript shows the claim. The conversation plan shows what prompted it. The supplied links show what informed it. The review step shows whether the final edited asset still matches the meaning of the answer.
This does not make every claim automatically true. A person can be wrong, vague, outdated, or overconfident. But it gives the editor somewhere to check the work. The alternative is often a polished draft with no memory of where the key line came from.
A fluent paragraph can hide thin evidence. A recorded answer preserves the speaker's reasoning, hesitation, examples, and limits for an editor to check.
REC prepares questions and organizes the material. The person supplies the answers and reviews any resulting clips.
What to record in the workflow
First, record the source material that shaped the conversation. Keep the links, documents, notes, product context, customer-safe examples, and public sources that informed the interview. If the prompt was built from research, keep the research.
Second, preserve the original answer. Keep the recording and transcript connected. If the transcript is corrected, keep enough history to recover the source when a claim is reused.
Third, label the AI role plainly. Did AI prepare questions, summarize research, clean punctuation, suggest highlights, draft a caption, or rewrite an approved answer into article form? Write that down in ordinary language.
Fourth, mark claim types before publishing. Separate first-hand experience, reported fact, analysis, opinion, product claim, customer language, and speculation. Each type needs a different review standard.
Fifth, approve the final asset against the source. The editor should ask whether the strongest sentence is supported by the recording, whether the caveat survived, whether the AI-assisted wording changed the meaning, and whether the named person would still stand behind it.
What teams should avoid
Do not let AI drafts become the source. A generated draft can help with structure, but it should point back to the interview, source document, or public record. If the generated draft is the only artifact left, authorship becomes hard to defend.
Do not treat a human approval click as the same thing as human authorship. Approval matters, but a person can approve wording that compresses, overstates, or invents more than intended. The review has to compare the final asset with the source.
Do not force a grand disclosure where a simple process note would do. A practical note such as "AI helped prepare questions and organize the transcript; the answers came from the recorded interview" is often more useful than broad language that makes every part of the work sound equally automated.
Do not publish claims that have no owner. If nobody can say where a claim came from, slow down. Ask the person again, find the source, narrow the wording, or cut the line.