Write and test ad-copy variants
A copy-testing plan for Google responsive search ads and Meta ad text: distinct message angles written to each platform’s limits, a fair test set up with the platform’s own testing tools, and a clear rule for calling a winner.
This is a practical, copy-paste process for you need new headlines and descriptions for google responsive search ads or new text for meta ads, and want to know which message works. Use it when you want a clear structure, then add your own details.
Verified on 2026-10-2816 sources cited
Quick start
1. Decide what you want to learn, for example "does price or reviews bring cheaper customers?". 2. Write a few different reasons to buy (price, speed, reviews, result), not the same line in new words. 3. Google: write 3 to 15 headlines of up to 30 characters and 2 to 4 descriptions of up to 90. Google mixes them in any order, so each line must make sense on its own. 4. Meta: keep the main text short, about 125 characters. To learn which message wins, put each message in its own ad and run an A/B test. If you put several text options in one ad, Meta decides who sees each one, so results per option are not a fair test. 5. Use the platform’s test tool so people are split fairly. Do not switch ads on and off by hand. 6. Give each test time: Meta recommends at least 7 days; for Google, a rule of thumb is at least 2 weeks. Pick the winner only if the test tool says the difference is significant (Google) or reaches the confidence level you set before launch (Meta). 7. Only claim what you can prove.
Before you start
- Conversions are not yet recorded in Google Ads or Meta; a copy test needs a recorded result to decide on. Set up conversion tracking first. Set up conversion tracking with GA4, Consent Mode and Meta CAPI
- You need copy for LinkedIn, Performance Max asset groups or Shopping; this method covers Google Search responsive search ads and Meta single image or video ads only.
- You want to test images or videos rather than words. Plan and refresh your Meta ad creatives
- The campaign is not built yet; set up keywords and structure first. You can still use this method’s writing rules for the first ads; skip the test until the ad group has about 4 weeks of tracked conversions (rule of thumb). Set up a Google Ads search campaign
- The ads are disapproved or not delivering; fix that before testing copy. Fix a disapproved or rejected ad
Full method
How to use this method
- 1. Copy or downloadCopy the prompt below, or download the method file and attach it to your chat.
- 2. Paste into any AIChatGPT, Claude, Gemini or Copilot. Replace each [PLACEHOLDER] with your details, or let the AI ask you.
- 3. Add your materialAttach your own files, data or brief so the AI works from your facts, not guesses.
The prompt
Paste it as-is. The AI will ask for anything still in [BRACKETS].
# Write and test ad-copy variants Method by Deepstory Research: deepstoryresearch.com/method/ad-copy-variant-testing You are a direct-response copywriter who writes and tests ad copy for Google Search and Meta. Follow the method below step by step. Goal: A copy-testing plan for Google responsive search ads and Meta ad text: distinct message angles written to each platform’s limits, a fair test set up with the platform’s own testing tools, and a clear rule for calling a winner. ## My inputs - Platform and campaign: [PLATFORM_AND_CAMPAIGN] (example: Google Search, teeth whitening ad group) - Offer and proof: [OFFER_AND_PROOF] (example: AED 799 whitening, 4.9 stars from 600 reviews) - Keywords or audience: [KEYWORDS_OR_AUDIENCE] (example: teeth whitening dubai, whitening price) - Current copy and results: [CURRENT_COPY_AND_RESULTS] (example: 6 headlines, Ad Strength Average, CTR 3.1%) - Words you cannot use: [WORDS_YOU_CANNOT_USE] (example: no "guaranteed", no "painless") Before you start: if any input above is still in [BRACKETS] or missing, ask me for it in one short numbered list and wait. Do not guess. If I attached files (data, briefs, Deepstory evidence files), treat them as the primary source. ## Steps to follow 1. Pick the one thing the test should answer and the metric that decides it. 2. Write 3 to 4 message angles (for example price, speed, proof, outcome) before writing lines. 3. Write copy to each platform’s limits and rules, checking every claim can be proven. 4. Set up the test with the platform’s tool so traffic is split fairly. 5. Decide in advance how long to run and what counts as a winner. 6. Read the result, keep the winner, and write the next angle to test. ## Deliver the output in this structure - Test question and deciding metric - Message angles - Google RSA headlines and descriptions (with pins, if any) - Meta primary text, headline and description options - Test setup - Winner rule and next test ## Quality rules - Every headline must make sense on its own and in any combination. - Count characters for every line and stay within limits. - Do not make claims (prices, ratings, "best", results) that the business cannot prove. - Label every number as FACT (with source and date), ESTIMATE (show the calculation) or ASSUMPTION. - If a Deepstory evidence file is attached, use its figures as the reference values and cite its source references; tag claims that are not from it as [UNVERIFIED]. - End with: the 3 assumptions that matter most, what would change the answer, and the single best next step.
What you need to give the AI
| Input | Placeholder | Example |
|---|---|---|
| Platform and campaign | [PLATFORM_AND_CAMPAIGN] | Google Search, teeth whitening ad group |
| Offer and proof | [OFFER_AND_PROOF] | AED 799 whitening, 4.9 stars from 600 reviews |
| Keywords or audience | [KEYWORDS_OR_AUDIENCE] | teeth whitening dubai, whitening price |
| Current copy and results | [CURRENT_COPY_AND_RESULTS] | 6 headlines, Ad Strength Average, CTR 3.1% |
| Words you cannot use | [WORDS_YOU_CANNOT_USE] | no "guaranteed", no "painless" |
Use it when
- You need new headlines and descriptions for Google responsive search ads or new text for Meta ads, and want to know which message works.
- You want to know which message works best on Meta: price, speed, proof or outcome.
- Ad Strength is "Poor" or "Average" and you do not know what to write.
Best with: Any chat AI; paste current ads, the keyword list or audience, and the last 30 days of ad or asset results.
What you get
- Test question and deciding metric
- Message angles
- Google RSA headlines and descriptions (with pins, if any)
- Meta primary text, headline and description options
- Test setup
- Winner rule and next test
The AI will ask you first
Answer these before the AI advises you; each one changes what it recommends.
What single question should this test answer, and which metric decides it?
A test with one variable and one decision metric gives a usable answer; Meta recommends testing one variable at a time.
If unknown: Test message angle; decide on cost per conversion. If each version is unlikely to reach about 30 conversions (rule of thumb), use a more frequent metric: on Google, conversions or conversion rate, whichever the test tool marks for significance (with a 50% split, total conversions compare fairly); on Meta, cost per a more frequent event closer to the sale (for example add to cart or lead). Use click-through rate only as a last resort, because a clickier headline can bring worse customers.
What is the offer, and which claims can you prove (price, ratings, guarantees, delivery times)?
Claims you cannot back up risk disapproval and complaints.
If unknown: Use only facts the user confirms; mark anything else NEEDS PROOF.
For Google: which keywords does the ad group target, what are the current Ad Strength and asset ratings, which bid strategy is used, and is AI Max or text customization on?
Ad Strength rewards keywords placed whole inside a headline; custom experiments need Smart Bidding; text customization adds Google-written lines that mix into a test.
If unknown: Assume text customization may be on; check the campaign settings and keep it the same in both versions.
For Meta: which placements run, and are Advantage+ text improvements or text generation turned on?
Text may be truncated by placement, and enhancements can change the text you wrote.
If unknown: Assume Advantage+ placements and check enhancements before publishing.
How many conversions does the campaign get a week, and how long does a typical customer take from first click to buying?
Low volume needs longer tests or a more frequent metric; Meta advises running longer than 7 days if customers usually take more than 7 days to convert.
If unknown: Fewer than 30 conversions per version a month and a buying time over 7 days: plan at least 6 weeks on Google and 14 days on Meta (rule of thumb), and check the first week’s volume before committing.
Are there words, claims or brand rules to avoid?
Regulated sectors and brand rules limit what copy can say; health, finance and cosmetics ads also have local rules (for example health advertising rules in Dubai, EU cosmetic claims).
If unknown: Avoid superlatives and outcome promises unless proven.
The procedure, step by step
Read this before you run it, so you can check the AI actually followed the method.
Set the question
One variable (angle, offer framing, call to action) and one deciding metric.
List angles
3 to 4 angles with the proof behind each.
Write Google lines
15 headlines of up to 30 characters (keywords whole inside a headline), 4 descriptions of up to 90; pin only mandatory text. In an angle test, use the shared lines plus 2 or 3 angle headlines per version instead.
Write Meta lines
Primary text about 125 characters or less, headline about 40, description about 25; one angle per ad when testing.
Check rules and claims
Editorial policy, Personal Attributes, proof for every claim, banned words list, and that the shared final URL shows every claim each version makes.
Set up the test
Google: an ad variation or a custom experiment, both of which split traffic for you (choose the cookie-based split in a custom experiment); a second responsive search ad in the same ad group is not a fair test. Meta: an A/B test with one variable, 7 to 30 days.
Run without edits
No other changes to the campaign during the test.
Decide and log
Winner by the pre-set rule; record angle, dates, result; plan the next test.
Platform settings(show)
| Platform | Path | Setting | Note |
|---|---|---|---|
| Google Ads | Campaigns > Ads > + > Responsive search ad | Headlines (up to 15, 30 characters), descriptions (up to 4, 90 characters), paths (15), pinning | |
| Google Ads | Ad editing panel | Ad Strength rating (Incomplete to Excellent) and action items | |
| Google Ads | Campaigns > Experiments | Custom experiment (cookie-based or search-based split); ad variations | |
| Meta Ads Manager | Ad > Primary text, Headline, Description > Add text options | Multiple text options; text generation (up to 5 variations) | |
| Meta Ads Manager | Toolbar > A/B test | Variable to test, winner metric, schedule 1 to 30 days |
Worked example
The figures below are illustrative. Use your own numbers and facts.
Scenario: A Dubai dental clinic’s Google Search ad group for teeth whitening.
Input: 6 headlines, all variations of "Teeth Whitening Dubai"; Ad Strength Average; CTR 3.1%; 40 conversions a month. Proof: AED 799 price, 4.9 stars from 600 Google reviews, 60-minute appointment. Avoid "guaranteed" and "painless".
Output excerpt: Question: does a proof message bring cheaper bookings than a price message? Deciding metric: cost per booking. Shared headlines (same in both versions): "Teeth Whitening Dubai" (21), "Whitening in One 60-Min Visit" (29), "Book Your Appointment Today" (27). Price headlines (original only): "Teeth Whitening From AED 799" (28), "AED 799 Teeth Whitening" (23). Proof headlines (variation only): "Rated 4.9 From 600 Reviews" (26), "4.9 Stars on Google Reviews" (27). Descriptions (same in both, with no price or rating so they do not leak an angle): "Teeth whitening in one 60-minute visit at our clinic." (53), "Book your teeth whitening appointment at our clinic today." (58). Pin the two angle headlines to Headline position 1 in each version, so every ad shown leads with its angle. They are close in wording, which Google says lowers Ad Strength when pinned to the same position; accept that during the test. Setup order: first check whether text customization (AI Max) is on for this campaign. If it is, Google-written lines can add price or rating wording to either version, so the user either turns it off for the test or accepts a blurred result and records it in the test log. Next, check that no sitelink, callout, price asset or other ad in this ad group mentions AED 799 or the rating, because those can show in both versions. Then edit the live ad into the price version (shared lines, price lines and the pins) and wait until it is approved, so the original side of the test serves the price version from the first day. Check that no sale or public holiday falls inside the 8 weeks; if one does, move the dates. Then create the variation with a start date 1 or 2 days ahead, so the new ads can be reviewed before it starts (rule of thumb). Test: an ad variation (Campaigns > Experiments > Ad variations) scoped to this campaign and filtered to this ad, using Update text to swap the price headlines for the proof headlines, with a 50% split; Google splits users by cookie. Do not use a second responsive search ad in the same ad group as the test: ad rotation favours the ad it expects to do better, so the split is not fair. Run 8 weeks: at 40 bookings a month that is about 74 bookings, about 37 per version; at this volume only a large difference will show as significant (rule of thumb). Call a winner only if the tool marks it significant; otherwise keep the price version and test a bigger change, such as a different offer. Check: the rating line must match the live Google rating on launch day and stay true for the whole run; check it weekly, and if the rating falls below 4.9, stop the test and fix the line (this is the one change that overrides the freeze). Health advertising rules in Dubai must be checked before reusing the copy, including whether "in one 60-minute visit" reads as a promise of immediate results.
Check your result
Before you trust the AI's output, confirm each of these.
- ☐ One test question and one deciding metric are written down.
- ☐ Every line is within the character limit and counted.
- ☐ Each claim has proof, and no banned word is used.
- ☐ The test uses the platform’s own split tool.
- ☐ Run time and winner rule were set before launch.
What you'll get(show template)
## Ad copy test: [AD GROUP / AD] Question: [ ] | Deciding metric: [ ] | Run: [dates] | Winner rule: [ ] ### Angles Angle | Proof | Example line ### Claims to confirm Claim | Proof | Status (confirmed / NEEDS PROOF) ### Google responsive search ad Headlines (chars): 1. [ ] ([n]) ... 15. [ ] Descriptions (chars): 1. [ ] ([n]) ... 4. [ ] Pins: [none / text and position] ### Meta text options Primary text (chars): [ ] | Headline: [ ] | Description: [ ] ### Test setup Google: [ad variation / custom experiment], split [%], same final URL in both | Meta: [A/B test, variable, audience, 7 to 30 days] ### Result and next test Winner: [ ] | Evidence: [ ] | Next test: [ ]
Quality checks built into the prompt
- • Every headline must make sense on its own and in any combination.
- • Count characters for every line and stay within limits.
- • Do not make claims (prices, ratings, "best", results) that the business cannot prove.
- • Every number labelled FACT, ESTIMATE or ASSUMPTION; claims outside attached evidence tagged [UNVERIFIED].
Decision rules
If/then rules the AI should apply. Heuristic rules are a practitioner's rule of thumb, not a platform rule.
| If | Then | Type |
|---|---|---|
| You write a Google responsive search ad. | Write at least 3 and up to 15 headlines (30 characters each) and at least 2 and up to 4 descriptions (90 characters each); paths are up to 15 characters, and characters in Chinese, Japanese or Korean count as 2. Google shows at least one headline and one description in any order, so each line must stand alone. | Platform rule |
| Some text must appear in every Google ad (for example a legal line). | Pin it to Headline position 1, Headline position 2 or Description position 1. Pin only what must appear: pinning reduces the combinations Google can test. | Platform rule |
| You plan the number of responsive search ads. | Google recommends (a best practice, not a requirement) at least 2 responsive search ads with "Good" or "Excellent" Ad Strength per ad group, each with a unique final URL. This is not a test: ad rotation set to Optimize favours the ad expected to do better, so two ads side by side are not a split test. When you test copy, keep the final URL the same in both versions. | Platform rule |
| Ad Strength is low. | Add more unique headlines and descriptions, put popular keywords whole inside a headline (not split across two), vary selling points and calls to action, and make sure 6 or more sitelinks apply. If text customization is on, Ad Strength also counts Google-generated text. Ad Strength does not decide whether an ad can serve; use it as guidance while writing. | Platform rule |
| You test one Google copy change (in one campaign or many). | Use ad variations (Campaigns > Experiments > Ad variations): choose all campaigns or one campaign, select the ads, then update text by adding, removing or pinning headlines and descriptions, and set a split; users are split by cookie. Read results: a blue asterisk means at least 95% likely the change caused the difference; the default confidence interval shown is 80%. | Platform rule |
| You test a bigger change in one Google campaign. | Use a custom experiment with a cookie-based or search-based split; the original campaign must stay active, and eligible Search campaigns use Smart Bidding with exact or phrase match keywords. Leave "Apply experiment changes ... if the treatment arm wins" off unless the user decides otherwise. For copy tests, choose the cookie-based split so each person sees only one version. Changes made to the original campaign during the test are not copied to the experiment, and only one experiment runs at a time. | Platform rule |
| Your Google Search campaign uses AI Max or text customization. | Google can generate and adapt headlines and descriptions, and Ad Strength counts generated assets as well as yours. Review generated text before the test, keep the setting identical for both versions, and record it in the test log. AI Max can also send people to other pages on your site (final URL expansion); if it is on, record it, because the landing page then varies as well. | Platform rule |
| Your Google copy uses capitals, repeated punctuation, symbols or a phone number in the text. | Check the Editorial policy first: incorrect capitalisation, gimmicky punctuation or symbols, repetition and phone numbers in ad text are disapproval reasons. The Repetition rule also bans unnecessary or gimmicky repetition, including assets that repeat text already in the ad text and asset text that repeats words or phrases from another asset in the same ad group, campaign or account; keep repeated wording to what the keyword needs (rule of thumb). | Platform rule |
| You write Meta ad text. | Keep primary text short (1 to 3 lines). Recommended lengths are about 125 characters for primary text, 40 for the headline and 25 for the description; longer text may be cut off. | Platform rule |
| You have several Meta text versions and want Meta to pick the best for delivery. | In a single image or video ad, add multiple text options for primary text, headline and description; the ad system chooses which to show each person. This optimises delivery; it is not a test. To learn which angle wins, put each angle in its own ad and use an A/B test. Text generation can suggest up to 5 AI variations of primary text and headline; review each before use. | Platform rule |
| You need a clean answer on Meta (for example price message versus proof message). | Use an A/B test, not switching ads on and off by hand: it splits the audience so nobody sees both versions. Change one variable, use an audience not used by other campaigns at the same time, set a budget large enough to produce results, and run 7 to 30 days; run longer than 7 days if customers usually take more than 7 days to convert. Give both versions the same budget for a fair comparison. If you create the test by duplicating a campaign, Meta duplicates its budget as well, so total spend rises. Before launch, check that Advantage+ text improvements are off, or set the same way in both versions, and check the ad account’s advertising settings: Meta says some enhancements are on by default, and if the account is opted in to test new optimizations, test enhancements may apply even when Advantage+ creative is off at the ad level. | Platform rule |
| You want to compare 2 to 7 Meta ad versions in one test. | Meta also offers a Creative Test: 2 to 7 ad versions, one per test cell, share a test budget you set, and the result names "high performers" (with a confidence value per ad) rather than one winner. Meta’s guide describes it for the Marketing API; where your account or tool offers it, it can test copy angles. Set the deciding metric, run length and test budget before launch, as for an A/B test. If you are unsure, use the A/B test. | Platform rule |
| Your Meta copy refers to the viewer. | Do not assert or imply their personal attributes ("Are you over 50?", "Struggling with debt?"). Speak about the product or the result. | Platform rule |
| You choose angles to test. | Test angles, not synonyms: price, speed, proof (reviews, numbers), outcome, risk reversal. Two headlines that say the same thing in different words teach you nothing. | Heuristic |
| The conversion the platform counts is not the business result (for example raw form fills when only some leads qualify). | Do not decide on cost per conversion alone: copy that attracts unqualified leads can win it. Pass the ad ID into your CRM, compare the qualified rate per version, and do not adopt a version whose qualified rate is clearly lower. | Heuristic |
| You test an angle in a Google responsive search ad. | Keep shared headlines and descriptions identical in both versions and free of either angle, put 2 or 3 angle headlines in each version, and pin those to Headline position 1. Ad Strength may read lower than a full 15-headline ad; accept that during the test. | Heuristic |
| You test an angle in Google ads. | Check the other ads in the ad group and the sitelinks, callouts, price assets and other assets that can show with the ad. They show in both versions, so any that carry one angle (a price, a rating) blur the test; pause or edit them before launch, or record them in the test log. Google can also show unused headlines from the same ad, or from another ad in the same ad group, as link-based assets. | Heuristic |
| A sale, holiday or seasonal peak falls inside the planned run. | End the test before the peak or start it after; do not change offer text during the run. | Heuristic |
| You read a Google ad variation or experiment. | Call a winner only when the tool marks the deciding metric as statistically significant (in ad variations, a blue asterisk means at least 95% likely the change caused the difference). Before launch, check that the tool reports significance for your deciding metric; if not, choose one it reports or treat the result as directional. | Platform rule |
| You read a Meta A/B test. | Meta names as winner the version with the lowest cost per result for the event you chose when creating the test, and shows a confidence percentage (the chance of similar results if the test were repeated). Meta warns that a winner can still have higher costs if the test was too short, had too few results, or did not follow its best practices, so check the run length and number of results before acting on it. If there is no clear winner, Meta may show "top performing versions" judged on other metrics; treat these as no winner. If the test did not collect enough data, Meta may recommend extending it. | Platform rule |
| You call a winner. | Decide in advance: the metric, the minimum run (at least 2 weeks on Google and at least 7 days on Meta, longer if a typical customer takes longer to buy), and the winner rule. On Meta, set the minimum confidence before launch (rule of thumb: 90% or more) and treat a lower figure as no winner. With fewer than about 30 conversions per version, only very large differences clear these bars, so plan a longer run or use a more frequent metric: on Google, conversions or conversion rate, whichever the test tool marks for significance (with a 50% split, total conversions compare fairly); on Meta, a more frequent event closer to the sale (for example add to cart or lead). If there is no winner, keep the original and test a bigger difference. | Heuristic |
Failure modes
Common mistakes with this method, how to spot them, and how to fix them.
| Mistake | Symptom | Fix |
|---|---|---|
| Writing 15 headlines that say the same thing. | Ad Strength stays low; no learning about messages. | Write by angle; vary selling points and calls to action. |
| Pinning every headline. | Google cannot test combinations; Ad Strength drops. | Pin only text that must appear. |
| Headlines that only work in order. | Odd or misleading combinations. | Each line must make sense alone. |
| Testing by switching ads on and off. | Overlapping audiences and unreliable results. | Use the platform test tools (ad variations, experiments, Meta A/B test). |
| Calling a winner after a few days. | The "winner" loses next month. | Pre-set the run time and significance rule. |
| Claims the business cannot prove. | Disapprovals or customer complaints. | Mark every claim with its proof; remove unproven ones. |
| Copy that speaks to the viewer’s condition on Meta. | Rejected under Personal Attributes. | Talk about the product and its result. |
| Google text customization adds its own headlines during the test. | Ads show lines you did not write; the angle test is mixed. | Check AI Max and text customization before launch, review generated assets, and keep the setting the same in both versions. |
| Putting several angles into each test version. | No angle wins; results are a blend. | One angle per version; shared lines identical in both. |
| Editing ads, budgets or settings while the test runs. | Results shift mid-test. | Freeze changes until the end date; log any unavoidable change. |
| Sitelinks, callouts, price assets or another ad in the ad group carry one angle. | Both versions show the price or rating; the difference shrinks. | Check assets and other ads before launch; pause or edit those that carry an angle, or record them. |
| Different final URLs in the two versions. | The winner may be the landing page, not the copy. | Use the same final URL in both versions. |
| Letting text improvements rewrite copy unchecked. | Published text you did not approve. | Review Advantage+ text enhancements before publishing. |
Agent mode: if an AI agent runs this unattended
May do without asking
- • Draft headlines, descriptions and Meta text options with character counts.
- • Read ad and asset results in a read-only session.
- • Check copy against the Editorial and Personal Attributes rules.
Must stop before (the user decides and does these)
- • Creating, editing or publishing any ad, which goes to review and can spend budget.
- • Changing bids, budgets, bid strategy, ad rotation or other campaign settings (for example switching to Smart Bidding so a custom experiment is allowed), or creating a Meta A/B test by copying a campaign, ad set or ad (copying a campaign also duplicates its budget, so total spend rises).
- • Starting or applying an experiment, ad variation, A/B test or Creative Test.
- • Turning on text generation, Advantage+ text enhancements, or Google AI Max or text customization.
- • Turning on "Apply experiment changes to your original campaign if the treatment arm wins", applying an ad variation, or accepting recommendations that change ad text.
- • Pausing, removing or editing ads or headlines in a running test, or ending a test early.
- • Accepting generative AI or advertising terms, or changing billing.
Evidence for this method
This method runs on your own material: your files, brief or code. If your task needs market or economic numbers, attach a sourced context file from the evidence library alongside the prompt.
Browse evidenceRun next
Methods that chain well after this one.
Deepstory methods are process guidance for AI tools. Review AI output before relying on it; legal, tax, medical and investment decisions need a qualified professional.