Claude Fable 5.2 Customer Research: Trace Each Ad Concept to Evidence

Quick answer

Number the supplied source excerpts, group recurring problems without claiming statistical prevalence, and request a concept with an evidence ID and a separate creative interpretation. Approve that interpretation before making visuals.

1. Build a small source packet

For a fictional commuter bag, prepare consented interview notes or public reviews you may use. Give each excerpt an ID, date and source. Remove personal information that the creative task does not need.

Keep negative and positive evidence together. Three complaints selected by a marketer do not establish what most buyers think. Record how the packet was chosen so the final concept is not presented as a representative survey.

2. Separate observations from the proposed message

Ask Fable to produce a theme, supporting IDs, counterexamples and a potential visual. A request for a pocket that fits a travel card can support a pocket-access concept. It cannot prove that the bag saves ten minutes each morning.

Use a separate column for claims awaiting product verification. Customer language about comfort does not establish an ergonomic performance claim. The concept should name an action that could be shown with the actual product.

3. Approve one concept and prepare a visual test

Choose the concept with a relevant problem and enough evidence to explain the choice. Write the exact product detail that must remain visible: pocket opening, card size or zipper position. Give the source photograph to the visual designer.

In Ciyo, manually request a concept image using the approved message and actual reference. Label it a visual proposal. Inspect whether the depicted feature exists on the real bag; a beautiful extra pocket invalidates the illustration rather than improving the product.

Copy and adapt this request

Replace the bracketed fields with your own approved information. The requested output is a draft for review; the example has not been run as a model benchmark.

From [SOURCE PACKET WITH IDS], group problems and desired outcomes. For each theme show supporting IDs, counterexamples and uncertainty. Propose three ad concepts with a visible product action, a distinct creative interpretation and a list of claims requiring verification. Do not invent quotes, prevalence, revenue or product features.

Evidence-to-concept map

These example checks describe acceptance criteria, not observed model results.

Common mistakes and how to recover

Do not count repeated copies of one review as independent evidence. Deduplicate the source packet first.

If the model merges conflicting themes, ask it to retain separate groups. A single slogan is not worth losing the distinction between different buyer needs.

Choose a working method and make the handoff

For a solo shop, start with one item and keep the source facts beside the draft. A team should name one approver and keep a dated version so an old comment cannot silently override the current decision. If the input is incomplete, return to the owner of that fact before polishing the artwork.

Use Ciyo for the visual work described here with the controls and models currently offered in your account. Keep the approved copy separately and add exact wording with editable canvas text where needed. A generated picture is a visual draft; inspect its exported file before using it.

• AI design agent
• Fable model overview

Sources and further reading

Primary documentation and related guides used for this article. Availability and controls can vary by account; verify them before beginning your own run.

• Anthropic model overview