Twelve points for an AI marketing tools quality checklist

What to take away

The commonest mistake is approving a tool on a vendor demo, then finding the quality problems after the licence is signed.
A quality checklist tests inputs, prompts, outputs, records and ownership, in that order, before anyone builds a campaign on the tool.
Every check needs a named owner and evidence you could show a client or an auditor.
Run the full twelve checks at onboarding, then a shorter version each quarter and after any model update.

Treat demos as a warning

A demo shows what the vendor's best operator can do with clean inputs and a prepared script. Your team will not work that way: quality failures start in the gap between the rehearsed example and routine output, so test the routine instead.

The checks below form the quality layer. The wider operations and delivery guide covers the budgeting and annual review cycle around them.

The twelve-point checklist

Work through them in order at onboarding. The first four cover inputs and prompts; points five to eight cover scoring, records and governance.

Define the campaign task and intended output. Evidencethe approved campaign brief. Owner: the campaign lead.
Record the input sources and data involved. Evidencean input inventory and representative sample. Owner: the data owner.
Test routine inputs, including incomplete or ambiguous briefs, rather than relying only on prepared demo inputs. Evidence: test inputs and resulting outputs. Owner: the tool evaluator.
Version approved prompts and record changes. Evidencea prompt register with version history. Owner: the prompt owner.
Set scoring criteria before testing, tied to the brief, such as accuracy, completeness and brand fit. Evidence: the agreed rubric. Owner: the campaign sign-off owner.
Score outputs against the rubric on the same ten briefs; record the result, reason and any correction. Evidence: the score sheet and reviewed outputs. Owner: the reviewer.
Decide what records to retain and document that decision. Evidence: the retention decision and a sample campaign file. Owner: the records owner.
Review whether the vendor uses sub-processors and what information they handle. Evidence: the review and its outcome, including if it is not applicable. Owner: the tool owner.

Twelve-point AI tool checklist

Confirm training data and prompt sources
Write down the single approved job
Test on ten real briefs
Score outputs against editor rubric
Check factual claims against primary source
Record prompt, model version, date
Confirm data access, storage, retention
Check vendor sub-processors and notice clauses

Prompts need a house standard

Prompt consistency is quality control, not a personal skill. If two marketers write prompts differently for the same task, you cannot compare outputs or trace what went wrong. Microsoft's Craft effective prompts with Microsoft 365 Copilot training path gives everyone a shared starting structure. Version the prompts that work, and retire the ones that produce inconsistent drafts.

Keep evidence you can show

Quality claims only hold up if you can produce the record. The ICO's guidance on accountability and governance expects organisations to demonstrate how they meet their data protection obligations. For a marketing team that means prompt logs, retention decisions and a named owner. If you could not produce that file in an afternoon, the checklist is not finished.

Who signs off what

Shared ownership quietly becomes no ownership. Name one person for the tool, one for the data question and one for published output, then write those names into the checklist. When the weekly rhythm is still unclear, the operating workflow for England teams sets out how briefs, drafts and sign-off move between people.

Budget and resourcing

Budgeting is separate from the quality checklist. The cost guide covers that question, but a cheap-looking tool can still fail on review time. A team paying £400 a month for a drafting tool that needs two hours of checking per campaign is not saving money.

Bringing in a freelancer to run prompt reviews raises tax and employment questions. The business and self-employed section on GOV.UK gathers starting guidance on hiring and paying people.

Common questions

How often should we repeat the checks?

Quarterly at minimum, and again whenever the vendor updates the underlying model. Re-run the scoring steps on the same ten briefs so you can compare like with like.

Do small teams need all twelve?

No. Keep the checks on inputs, scoring, records and ownership. Drop the sub-processor review if you run a single seat and handle no client data.

Who should own the checklist?

Usually the person who already owns campaign sign-off. Split it across a committee and the checks stop happening within a couple of months.

What counts as evidence?

The prompt, the output, the correction, the date and the reviewer's name. Store them with the campaign file.