The 12 CCAO-F anti-patterns: wrong answers, catalogued

Twelve wrong answers that recur across the CCAO-F question bank. Learn the shape and you can eliminate one or two options before you finish reading them.

1. Treating fluent output as verified output

Wrong: Sending a well-written response because it reads authoritatively.

Right: Fluency is a property of the writing, not evidence about the facts. Check the specific claims — figures, citations, names, dates — against a source.

2. Disclosing AI involvement instead of getting it reviewed

Wrong: Adding "prepared with AI assistance" to a document in a regulated domain and sending it.

Right: In legal, medical, financial, employment, housing, insurance, academic and journalistic work, disclosure and qualified human review are both required. Neither substitutes for the other.

3. Regenerating instead of diagnosing

Wrong: Asking again, and again, hoping for a better roll.

Right: Identify what was missing — audience, format, source material, a constraint — and supply it. The same prompt produces the same class of answer.

4. Reaching for a more capable model to fix a prompt problem

Wrong: Escalating to the heaviest tier when output is generic.

Right: Generic output is nearly always missing context, not missing capability. Fix the prompt first; change tier only when the task genuinely exceeds the one you chose.

5. Pasting confidential data because the tool is "approved"

Wrong: Treating an approved vendor as blanket permission for any content.

Right: Tool approval and data classification are separate questions. Check what your policy permits for this kind of information, and minimise or anonymise what you paste.

6. Letting Claude make a decision about a person

Wrong: Acting on a ranked shortlist of candidates, a lending recommendation, or a suggested diagnosis.

Right: Use it to prepare the analysis. A qualified human makes and owns any decision affecting someone's job, money, health or legal standing.

7. Uploading everything to Project knowledge

Wrong: Treating a Project as an archive, including superseded versions.

Right: Curate it. Stale documents get used confidently and produce outdated output. Remove the old version rather than instructing Claude to ignore it.

8. Re-explaining the same context every week

Wrong: Starting each recurring task from a blank chat and retyping the background.

Right: Configure it once. Recurring work with stable background material is exactly what Projects exist for.

9. Asking one prompt to do four jobs

Wrong: "Analyse this report and write a summary, an email and a slide outline." Everything comes back shallow.

Right: Sequence it, with a check between steps, so an error in the analysis is caught before it propagates into three deliverables.

10. Overselling it to colleagues

Wrong: Promising a specific productivity percentage, or implying review is no longer needed.

Right: Concrete about what it does well, explicit about what needs checking and by whom, honest about what it cannot reach or guarantee.

11. Applying the same level of scrutiny to everything

Wrong: Either checking nothing, or subjecting an internal brainstorm to full source verification.

Right: Scale the check to the consequence of being wrong. Disproportionate process is a wrong answer too — it is the reason good workflows get abandoned.

12. Using Research when you needed Projects, or the reverse

Wrong: Running a research report to answer a question about your own attached documents; or setting up a Project to answer a one-off question about current events.

Right: Research is for current information from many external sources with citations. Projects are for recurring work over stable material you supply.

← Previous: Domain 7 — Troubleshooting and Optimisation  |  Next: Cheat sheet →