Build a paid social creative testing system from one campaign message

Start in Airtable and create one control record for the campaign before generating any creative. Add fields for campaign name, target audience, buyer problem, offer, approved proof point, landing page, CTA, primary test variable, variables to keep fixed, review owner, and approval status. Write the campaign message as one sentence using this structure: audience, problem, offer, proof, and desired action. Mark the primary test variable as message angle, visual format, offer framing, or proof type so the team knows what the test is actually measuring. Before moving on, have the campaign owner approve the message and reject any version that tries to test multiple audiences, offers, and creative formats at once.

Create a second Airtable table for the source material every creative variant is allowed to use. Add fields for approved claim, proof source, customer example, screenshot link, product area, banned wording, required disclaimer, landing page promise, brand rule, and claim approval owner. Include prior ad performance if you have it, but keep it tied to the original audience, offer, and channel so old results are not overgeneralized. Add a view called Needs approval that only shows claims without approval status and a view called Ready for creative that only shows claims approved for use. Do not let Claude, Canva, or AdCreative.ai invent proof that is not already in this table.

Use Claude to turn the approved campaign message into six to eight creative hypotheses, then paste the output into an Airtable hypothesis table. Create fields for hypothesis name, buyer belief, creative angle, required proof, visual direction, CTA, expected behavior, primary metric, risk, and review status. Each hypothesis should test a real belief, such as whether buyers respond more to risk reduction, speed, competitive pressure, peer proof, cost control, or missed-opportunity framing. Reject hypotheses that are only copy variations, because they will not teach the team anything meaningful. QA the set by checking that every hypothesis uses the same audience, offer, and landing page unless the campaign is explicitly designed to test one of those variables.

Create paid social creative hypotheses from this B2B campaign message.

Campaign message:
{{campaign_message}}

Target audience:
{{target_audience}}

Offer:
{{offer}}

Approved proof points:
{{approved_proof_points}}

Landing page promise:
{{landing_page_promise}}

Prior performance notes:
{{prior_performance_notes}}

Rules:
- Keep audience, offer, and landing page fixed unless I explicitly say otherwise
- Do not invent claims or proof
- Do not create slight wording variations as separate hypotheses
- Make each hypothesis falsifiable

Create 6-8 hypotheses. For each include:
1. Hypothesis name
2. Buyer belief being tested
3. Creative angle
4. Required proof
5. Visual direction
6. CTA
7. Expected behavior if the hypothesis is true
8. Primary metric
9. Risk or weakness
10. Reject/revise notes if the hypothesis is not testable

Ask Claude to create one or two ad concepts for each approved hypothesis, then add them to an Airtable concepts table. Use fields for concept name, linked hypothesis, primary text, headline, visual idea, template type, CTA, landing page alignment note, claim that needs approval, and concept status. Keep the test balanced: do not create eight concepts for one hypothesis and one concept for another unless you intentionally want to weight the test. Review the table for hidden variable changes, such as one concept using a stronger CTA, different offer, or different audience language. Approve only concepts where the hypothesis, copy, visual direction, and landing page promise clearly match.

Turn these creative hypotheses into a balanced paid social ad concept matrix.

Approved hypotheses:
{{approved_hypotheses}}

Brand voice:
{{brand_voice}}

Channel:
{{channel}}

Landing page:
{{landing_page}}

Source pack:
{{source_pack}}

For each concept, output:
1. Linked hypothesis
2. Concept name
3. Primary text
4. Headline
5. Visual idea
6. Template type
7. CTA
8. Landing page alignment note
9. Claim that needs approval
10. Risk of confounding the test

Rules:
- Keep the test balanced across hypotheses
- Do not change offer or audience unless that is the stated test variable
- Flag unsupported claims instead of rewriting around them

In Canva, create a small template set before producing individual ad files. Build templates for stat card, pain-point card, proof card, product screenshot card, comparison card, and quote-style card, then lock the core brand elements where possible. Create placeholders for headline, short proof line, CTA, product image, logo placement, and disclaimer so variants can be produced consistently. If you are using Canva Bulk Create, prepare a spreadsheet with one row per variant and columns that match those placeholders. QA the template set by exporting one sample of each format and checking mobile readability, safe margins, contrast, visual hierarchy, and whether the message is understandable in three seconds.

Use Canva and AdCreative.ai to produce a controlled batch of variants from the approved concept matrix. Create fields in Airtable for variant ID, linked concept, linked hypothesis, template type, copy version, visual version, proof source, file link, export size, platform placement, and review status. Use a naming convention that exposes the test logic, such as campaign, audience, hypothesis, format, and version. Generate more variants than you plan to launch, but only move the clearest ones into review. Before approval, confirm that every file can be traced back to a hypothesis, concept, proof source, and landing page promise.

Create an approval view in Airtable that shows only variants with missing review status, unsupported claims, or landing page mismatches. Review each ad for claim accuracy, source support, brand fit, mobile readability, visual-message match, CTA clarity, and whether the landing page delivers the same promise. Use Claude as a second-pass reviewer, but keep the final decision with a human owner. Mark each variant as approved, revise, or reject, and require a revision reason for anything that does not pass. Do not launch variants with vague proof, unclear audience fit, unreadable mobile text, or a promise the landing page does not immediately support.

Review these paid social creative variants against our QA criteria.

Creative variants:
{{creative_variants}}

Brand rules:
{{brand_rules}}

Approved claims:
{{approved_claims}}

Landing page promise:
{{landing_page_promise}}

Target audience:
{{target_audience}}

For each variant, assess:
1. Claim accuracy
2. Proof support
3. Brand fit
4. Mobile readability
5. Visual-message match
6. CTA clarity
7. Landing page alignment
8. Approve, revise, or reject
9. Specific revision note

Be strict. Do not approve vague, unsupported, unreadable, or mismatched creative.

Create the final launch matrix in Airtable before uploading anything to the ad platform. Add fields for audience, platform, campaign, ad set, creative variant, linked hypothesis, format, CTA, landing page, budget, start date, end date, primary metric, secondary metric, UTM link, and launch status. Keep the audience, placement, budget, and landing page consistent enough that the creative hypothesis can still be interpreted. For LinkedIn or another paid social platform, use campaign and ad names that match the Airtable variant ID so reporting can be joined later. QA the matrix by filtering for blanks in linked hypothesis, UTM link, landing page, approval status, and primary metric before launch.

Use Looker Studio, platform reporting, or your paid media export to review results by hypothesis instead of only by ad name. Track impressions, CTR, CPC, conversion rate, CPL, qualified conversion rate, spend, frequency, comments, and any sales-quality notes available from the CRM. Add calculated fields or grouped dimensions so each variant rolls up to its linked hypothesis, format, and proof type. Do not judge too early if spend or impressions are too low, but do flag obvious issues like high CTR with no conversions, low CTR with strong conversion quality, or a proof angle that attracts unqualified traffic. Add a performance note to Airtable for every hypothesis, not just the winners.

Feed Claude the hypothesis matrix, launch matrix, performance readout, QA notes, and qualitative comments. Ask it to summarize which beliefs look stronger, which formats created misleading engagement, which variants should be paused, and what the next test should isolate. Add the final learning brief back into Airtable with fields for decision, next action, owner, deadline, and reuse status. Keep winning templates and proof angles, but retire weak directions so the next sprint does not repeat the same mistakes. Close the loop by updating the source pack with new rules, such as claims that performed well, formats that confused buyers, or landing page promises that need better support.

Summarize paid social creative learnings from this test.

Hypothesis matrix:
{{hypothesis_matrix}}

Launch matrix:
{{launch_matrix}}

Performance readout:
{{performance_readout}}

QA notes:
{{qa_notes}}

Qualitative comments:
{{qualitative_comments}}

Output:
1. Strongest hypotheses
2. Weakest hypotheses
3. Best creative formats
4. Message-audience fit observations
5. Landing page mismatch risks
6. Variants to pause
7. Variants to iterate
8. Claims or proof points to reuse
9. Next test recommendation
10. One executive summary paragraph

Separate data-backed observations from speculation.