Track each photograph and its caption as separate content records linked to one figure. An author-owned photograph does not make an AI-written caption author-written, and an accurate caption does not establish the image rights. Review origin, factual accuracy, and the intended publication use separately before assembling the final figure.
Give the photograph and caption separate identities
Assign a stable identifier to the photograph and connect it with a caption record. Preserve the image source, creator information, and relevant publication-use records beside the image. Preserve the caption's drafting history and approved wording beside the text.
This distinction matters when one photograph appears in several places. An overview caption and a detail caption may explain different things, even when they use the same image source. Keep each caption connected to its actual placement so later edits do not replace one explanation with another merely because the filenames match.
Record what the AI tool actually contributed
Note whether the tool created the caption text, suggested ideas, or edited wording the author had already written. KDP distinguishes AI-generated content from AI-assisted work based on the role of the tool. Substantial editing does not automatically change the origin of text that the tool created.
Keep the original generated output when it supports the editorial record, but clearly identify the version retained in the manuscript. A folder of unused caption candidates should not be confused with the final figure text. The record needs to explain what actually entered the publication and how it was prepared.

Verify the visible description against the image
Read the proposed caption while inspecting the exact photograph selected for use. Check objects, positions, quantities, and any labels that the text identifies. A generated description may sound plausible while referring to a detail that is not visible or that belongs to a different image.
Avoid inferring more than the photograph shows. A caption may identify a tool on a table without establishing who used it, when the work happened, or why it was placed there. Those additional claims need their own evidence rather than being accepted because they fit the surrounding narrative.
Check contextual claims against source records
Verify dates, locations, names, and explanations using the relevant records. If the photograph belongs to a documented event, connect the caption with that evidence. If the information is uncertain, preserve the uncertainty in the editorial record and use appropriately limited wording in the final text.
Review any practical instruction the caption adds. A photograph of a process does not automatically validate a generated recommendation about how to perform it. The caption should agree with the reviewed manuscript and should not introduce new instructions that bypass the book's normal content review.

Recheck captions after image changes
When a photograph is replaced, cropped, or retouched, inspect the associated caption again. A detail named in the text may have disappeared from the frame, and a directional phrase may no longer point to the intended object. Mark the figure for review when the visual change affects what the caption describes.
Keep labels and figure references synchronized as well. If the image number changes after reordering, confirm that the chapter points to the correct photograph and that the caption follows it. Origin tracking is useful only when the record remains attached to the actual figure used in the manuscript.
Reconcile the final figure with the publication record
Before export, confirm the approved photograph, crop, caption, and placement as one assembled figure. Then inspect the exported page for readable text and an unambiguous relationship between the image and its explanation. Preserve the separate source records behind that finished presentation.
Use the retained caption history when reviewing the book's AI-content disclosure, alongside the records for other text, images, and translations. Keep that origin review distinct from factual and rights checks. Each answers a different question, and completing one does not automatically complete the others.
Frequently asked questions
Clear answers for this publishing decision.
Does taking the photograph myself make an AI-generated caption AI-assisted?
No. Record how the caption text itself was created. The origin of the photograph and the origin of the accompanying text are separate facts.
Can a caption include information not visible in the image?
It can provide verified context, but that information needs an appropriate source. Do not treat a generated guess about a date, person, or event as evidence supplied by the photograph.
What should happen when I crop the image?
Recheck the caption, labels, and surrounding references against the new view. The words may still be correct, but approval should reflect the version readers will actually see.
What must I check, disclose, and verify myself when AI helped produce the book?
Review facts, rights, and source material yourself, record which parts were AI-assisted or AI-generated so you can answer Amazon’s disclosure question accurately, and treat every automated check as a starting point rather than a verdict.
See what the compliance checks cover