Optimization

Test validity check: were the results contaminated by mid-test changes

A read-only audit of Google Ads and Meta change history, experiments and conversion tracking during a test window, ending with a Valid, Valid with caveats or Rerun needed verdict per test.

When to use it

When you are about to report A/B test results to a client and need to know if changes made mid-test, such as budget, bid or tracking edits, skewed the outcome.

What you get

A timeline table of every change that could bias each test arm, plus a Valid, Valid with caveats or Rerun needed verdict per test and a list of blind spots.

Does it change anything in my account?

No. The assistant only reads your data and reports back.

Works with

Google Ads, Meta Ads

Fill in before sending

[client name][start date][end date][Google Ads account][Meta ad account]

Before I report test results to client [client name], check whether the tests were contaminated. This is a read-only review. Make no changes. Test window: [start date] to [end date]. Google Ads account: [Google Ads account], Meta ad account: [Meta ad account]. Tests and arms: [test name: arm A = campaign or ad set name/ID, arm B = campaign or ad set name/ID; repeat per test]. 1. Confirm the accounts with list_accounts and list_meta_ad_accounts, and match each arm I listed to its real ID with list_meta_campaigns and list_meta_adsets (for Google, use the IDs that appear in the change history). If an arm name does not match anything, stop and ask me. 2. Google: pull get_change_history. It only reaches back 30 days, so if [start date] is older than that, tell me which part of the window you could not see. Filter the rows yourself to the test dates and the arm campaigns. Also run list_experiments (action list) and note any experiment that overlapped the window or locked an arm. 3. Meta: pull meta_change_history with since and until set to the test window, then keep only events on the arm campaigns, ad sets and ads. 4. List every change that could bias a result: - budget changes above 20% - bid strategy or target CPA/ROAS changes - ads added, paused or edited - keyword, negative keyword, audience or targeting edits - conversion action or event changes 5. Tracking: run conversion_tracking_health and list_conversion_actions for the window on Google. On Meta, check list_meta_pixels (last fired) and list_meta_custom_conversions (never fired). A paused or silent conversion action must not be mistaken for a losing variant. Meta activity history does not show every pixel or event change, so if tracking looks off but no matching change is logged, say it could not be traced. Output a timeline table with these columns: date, platform, user, entity, change, old value, new value, affected test arm, and impact (High, Medium or Low). High means the change hit only one arm, or it touched budget, bidding or conversions. Account-level changes affect all arms equally; mark them as such. End with a verdict per test: Valid, Valid with caveats, or Rerun needed, with one line of reasoning each. List any blind spots (days outside the Google 30-day history, untraceable tracking changes) under the verdicts.

Before I report test results to client [client name], check whether the tests were contaminated. This is a read-only review. Make no changes.

Test window: [start date] to [end date].
Google Ads account: [Google Ads account], Meta ad account: [Meta ad account].
Tests and arms: [test name: arm A = campaign or ad set name/ID, arm B = campaign or ad set name/ID; repeat per test].

1. Confirm the accounts with list_accounts and list_meta_ad_accounts, and match each arm I listed to its real ID with list_meta_campaigns and list_meta_adsets (for Google, use the IDs that appear in the change history). If an arm name does not match anything, stop and ask me.

2. Google: pull get_change_history. It only reaches back 30 days, so if [start date] is older than that, tell me which part of the window you could not see. Filter the rows yourself to the test dates and the arm campaigns. Also run list_experiments (action list) and note any experiment that overlapped the window or locked an arm.

3. Meta: pull meta_change_history with since and until set to the test window, then keep only events on the arm campaigns, ad sets and ads.

4. List every change that could bias a result:
- budget changes above 20%
- bid strategy or target CPA/ROAS changes
- ads added, paused or edited
- keyword, negative keyword, audience or targeting edits
- conversion action or event changes

5. Tracking: run conversion_tracking_health and list_conversion_actions for the window on Google. On Meta, check list_meta_pixels (last fired) and list_meta_custom_conversions (never fired). A paused or silent conversion action must not be mistaken for a losing variant. Meta activity history does not show every pixel or event change, so if tracking looks off but no matching change is logged, say it could not be traced.

Output a timeline table with these columns: date, platform, user, entity, change, old value, new value, affected test arm, and impact (High, Medium or Low). High means the change hit only one arm, or it touched budget, bidding or conversions. Account-level changes affect all arms equally; mark them as such.

End with a verdict per test: Valid, Valid with caveats, or Rerun needed, with one line of reasoning each. List any blind spots (days outside the Google 30-day history, untraceable tracking changes) under the verdicts.

Open in Claude Open in ChatGPT Discuss in the forum

More prompts

OptimizationChanges after your approval

Find search terms that waste budget and turn them into negatives

Combines smart negative suggestions with the full search terms report to flag zero-conversion and overpriced terms, then previews a grouped batch of negatives that protects your brand and converting keywords.

When to use itWhen your Google Ads search terms report shows money going to irrelevant or expensive searches and you want to cut that waste safely.

Review the search terms in my Google Ads account [account name] for the last 30 days. Use the smart negative keyword suggestions and the full search terms report together. My brand name and its common spellings are: [brand terms].

Flag two groups:
1. Terms that spent more than [spend threshold] with zero conversions.
2. Terms with exactly one conversion where the cost of that term is above 2x my target CPA of [target CPA]. Calculate this yourself from the cost and conversions in the search terms report.

Skip any term the report lists under already_negative, and check my existing negatives (including shared lists) so you never propose a duplicate.

For the matched keyword, query search_term_view with the keyword segment. If that field does not come back for a term, write "not available" instead of guessing.

Give me one table sorted by cost, highest first. Columns: search term, campaign, ad group, matched keyword, clicks, cost, conversions, recommended negative, match type, level (campaign or ad group).

Decision rules:
- Use exact match for one-off variants.
- Use phrase match when the same unwanted word appears in several terms, for example "free", "jobs" or "DIY".
- Pull my converting keywords from the keyword report. Never suggest a negative that would block one of them or any of my brand terms.
- If a term is relevant but simply expensive, list it separately as a bid or ad copy issue, not as a negative.

End with the estimated monthly savings. Use the last 30 days of cost from the terms you would block as the estimate, and say that this is an upper bound. Then show a preview of the negatives you would add, grouped by campaign, as one batch per campaign. Wait for my approval before adding anything.

OptimizationChanges after your approval

Block searches for places outside your service area

A list of search terms mentioning locations you do not serve, the location setting that lets them in, and a previewed fix for both.

When to use itWhen you suspect your ads are showing and costing money for searches from cities or areas you do not serve.

OptimizationChanges after your approval

Month-end budget pacing check across all client accounts

Runs pacing on every connected client account, compares projected month-end spend with each client's agreed monthly cap, and proposes CPA-ranked daily budget changes within guardrails, approved account by account.

When to use itWhen you manage several client accounts and need to know before month-end who is overspending or underspending against the agreed budget.

List my connected Google Ads client accounts (include client accounts under my manager account) and run a budget pacing check on each one for the current month, passing each account's id directly. The monthly caps agreed with each client are: [account name: monthly budget list]. Skip any account that is not on this list and tell me which ones you skipped.

For each account, take month-to-date spend per campaign from a campaign-level performance report for this month, and the daily budget and pace from the pacing check. Calculate projected month-end spend as spend to date plus current daily budget times the days remaining. Add the campaigns up to get the account total and compare it with the client's cap.

Deliver one table sorted by risk. Include these columns:
- account
- campaign
- daily budget
- spend to date
- projected month-end spend
- account variance vs the monthly cap, as an amount and as a %
- flag

Use these flag rules, at account level:
- OVER: projected spend is more than 5% above the cap.
- UNDER: projected spend is more than 10% below the cap.
- ON TRACK: everything else.

For every campaign in a flagged account, add month-to-date conversions and cost per conversion. This shows whether the underspend is worth fixing.

Then, for each flagged account, propose new daily budgets based on the days remaining. The goal is to land total spend within ±3% of the cap by month-end. When adding budget, prioritise the campaigns with the lowest cost per conversion. When cutting, start with the highest. Respect the budget guardrails: no single increase above +50%, no more than +100% within 24 hours, no cut deeper than -80%. If the target cannot be reached within those limits, say so and suggest a second step later in the week or a pause instead.

Show the proposal as a before/after table per account, then a preview of the changes. Wait for my approval account by account before changing any budget. Do not touch bidding strategies or targets.

OptimizationChanges after your approval

Shift budget from rank-limited to budget-limited Search campaigns

A budget-neutral reallocation plan that moves money to Search campaigns losing impression share to budget at an acceptable CPA.

When to use itWhen some Search campaigns run out of budget and miss impressions while others underspend or cost too much per conversion.

In [client account], analyse all Search campaigns for the last 30 days. Target cost per conversion for this client is [target CPA].

1. Run an impression share analysis. Put each campaign in one bucket: mainly losing to budget, mainly losing to rank, or not limited.
2. Add cost, conversions, cost per conversion and current daily budget for each campaign.

Decision rules:
- A campaign RECEIVES budget only if both are true:
  - search impression share lost to budget is above 10%
  - cost per conversion is at or below target
- A campaign DONATES budget if either is true:
  - it spends less than 85% of its daily budget
  - its cost per conversion is more than 30% above target
- Rank-limited campaigns get no extra budget. List them separately. Use keyword Quality Score data to say whether the main fix is bids or Quality Score.

Deliver:
- a table of all campaigns with bucket, metrics and verdict (receive / donate / hold)
- a reallocation plan that keeps total daily budget unchanged, with no single campaign changed by more than 30%

Show the before/after budgets. Wait for my approval before applying any change.