Marketing · Advertising
The Best Prompt for Ad Copy
The best prompt for ad copy produces variations you can test and claims you can back up. This prompt has the model pick four different angles before writing, then deliver three variations per angle with character counts for your platform's format. It limits factual claims to your own proof points, checks the promise against your landing page, and ends with a claim-check table and a simple plan for which angles to test first.
Ad Copy Prompt
Generates ad variations across four distinct angles, counted against your character limits, with every claim traced to proof you supplied.
- Works with
- ChatGPT, Claude, Gemini
- Difficulty
- Intermediate
- Complexity
- Multi-part reply
- Variables
- 7
- Version
- 1.0
Customize this prompt
Fill in what you know. Anything left blank stays as a [bracketed placeholder]. Nothing you type leaves your browser.
7 variables · fields marked * matter most
You are a performance copywriter. Write ad copy that can be defended claim by claim and tested angle by angle. ## What I'm advertising Platform and placement: [Platform and placement] Product or service: [Product or service] Audience: [Audience] Offer and the action I want: [Offer and desired action] Proof points I can substantiate: [Proof points you can substantiate] What the landing page says: [What the landing page says] Format spec and character limits: [Format spec and character limits] ## How to work First, propose four angles that differ in substance, not wording. Draw them from this set: the problem the audience has, the outcome they want, a specific proof point, an objection answered, a comparison with what they do now. For each angle write one line on which part of the audience it speaks to and why. Then write the copy. For each angle give three variations of every text component the format needs (for example headline, primary text, description, call-to-action label). Show the character count in parentheses after every line that has a limit. ## Hard rules - Stay within the limits in the format spec. If I left the spec blank, state the limits you are assuming at the top and tell me to check them against the platform's current documentation. - Every factual claim must trace to a proof point above. No invented numbers, rankings, customer counts, reviews, awards or guarantees. - No unsupported superlatives ("best", "#1", "fastest") and no promised results. - Do not create urgency or scarcity that is not in the offer. - The headline promise must match what the landing page delivers. If the landing page summary does not support an angle, say so instead of writing it. - Address the reader's situation without asserting personal attributes about them, such as health, finances or identity. - Flag any line that touches a regulated or sensitive area with [POLICY CHECK] so I can review it against the platform's advertising policies. - Write the way the audience talks. Cut filler words; put the concrete detail first. ## Output 1. Angle table: angle, audience slice, rationale. 2. Copy grouped by angle, each variation on its own line with its character count. 3. Claim check: a table listing each factual claim used and the proof point it relies on. 4. Test plan: which two angles to run against each other first, the single variable that differs between them, and what result would tell me to move on to the next angle. Do not predict click-through or conversion rates.
Use it when you are
- Drafting headline and description sets for a search campaign
- Producing copy variations for paid social tests
- Refreshing ads that have been running unchanged for months
- Giving a client or manager several angles to choose from before spending on creative
Why this prompt works
Angles are chosen before words
Ask for ten headlines and you usually get one idea phrased ten ways. Requiring four angles that differ in substance, each with a rationale, means the variations give you something real to test: whether the audience responds to the problem, the outcome, the proof or the answered objection.
You supply the limits, the model counts against them
Platform specs change, so the prompt takes limits as an input instead of trusting what a model remembers. Where you leave the field empty, the model has to state what it assumed. Counts shown after every line make over-length copy easy to spot.
The claim check makes review fast
Advertising claims carry legal and platform-policy risk. Restricting facts to your proof points and listing each claim beside its source in a table lets you, a client or a compliance reviewer verify the copy in a minute instead of reading every variation with suspicion.
Message match is tested inside the prompt
An ad that promises something the landing page does not deliver wastes the click. Supplying a summary of the page lets the model reject angles the page cannot support, which is a check copywriters normally do by hand. It is cheaper to drop an angle here than after paying for the traffic.
It ends with a test, not a forecast
The closing test plan names two angles, the single variable between them and a decision rule, and it bars predicted rates. That keeps the output honest about what copy can be known to do before it runs. What you learn from the first comparison then decides which angle runs next.
How to use it
- Pick the platform, then paste the current character limits and format requirements from that platform's help documentation.
- List only proof points you could show evidence for if asked. Include the exact terms of any offer and its real end date.
- Summarize or paste the landing page headline and main sections so the model can match the promise.
- Read the angle table first and replace any angle that does not fit before reviewing the copy.
- Verify character counts in the ad platform, review every [POLICY CHECK] line against the platform's ad policies, and confirm the claim-check table.
- Launch the two suggested angles, changing only the variable named in the test plan.
AI model compatibility
| Model | Status | Notes |
|---|---|---|
| ChatGPT | Expected to work | Character counts produced by a language model can be off by a few characters. Paste the final lines into the ad platform or a character counter before you rely on them. |
| Claude | Expected to work | If you have existing ads and their results, paste them in <current_ads> tags and ask for angles that differ from what is already running. |
| Gemini | Expected to work | State limits as explicit numbers in the format spec field, for example "headline: 30 characters", so the instruction is unambiguous. |
“Expected to work” means the prompt uses plain, model-independent structure and has been reviewed for that — not that it has been run on that model. A model is marked Tested only when a test record is published below.
Worked example
Scenario: An illustrative mobile car detailing service running a search campaign with a first-booking offer.
You are a performance copywriter. Write ad copy that can be defended claim by claim and tested angle by angle. ## What I'm advertising Platform and placement: Google Search Product or service: Brightline, mobile car detailing that comes to your home or office in Austin Audience: Busy professionals and parents who own a car less than six years old and dislike waiting at a car wash Offer and the action I want: Book a full interior and exterior detail online; first booking $20 off through the end of the month Proof points I can substantiate: We bring our own water and power; a full detail takes about 2.5 hours; fully insured; bookings available seven days a week What the landing page says: Headline: "Car detailing that comes to you." Shows three packages with prices, a service-area map and an online booking calendar. Format spec and character limits: Headlines: 30 characters. Descriptions: 90 characters. ## How to work First, propose four angles that differ in substance, not wording. Draw them from this set: the problem the audience has, the outcome they want, a specific proof point, an objection answered, a comparison with what they do now. For each angle write one line on which part of the audience it speaks to and why. Then write the copy. For each angle give three variations of every text component the format needs (for example headline, primary text, description, call-to-action label). Show the character count in parentheses after every line that has a limit. ## Hard rules - Stay within the limits in the format spec. If I left the spec blank, state the limits you are assuming at the top and tell me to check them against the platform's current documentation. - Every factual claim must trace to a proof point above. No invented numbers, rankings, customer counts, reviews, awards or guarantees. - No unsupported superlatives ("best", "#1", "fastest") and no promised results. - Do not create urgency or scarcity that is not in the offer. - The headline promise must match what the landing page delivers. If the landing page summary does not support an angle, say so instead of writing it. - Address the reader's situation without asserting personal attributes about them, such as health, finances or identity. - Flag any line that touches a regulated or sensitive area with [POLICY CHECK] so I can review it against the platform's advertising policies. - Write the way the audience talks. Cut filler words; put the concrete detail first. ## Output 1. Angle table: angle, audience slice, rationale. 2. Copy grouped by angle, each variation on its own line with its character count. 3. Claim check: a table listing each factual claim used and the proof point it relies on. 4. Test plan: which two angles to run against each other first, the single variable that differs between them, and what result would tell me to move on to the next angle. Do not predict click-through or conversion rates.
What a good response should contain
- Four clearly different angles, for instance time saved, convenience of on-site service, the insured and self-contained setup, and the first-booking offer
- Three headline and description variations per angle, each followed by a character count within the supplied limits
- No claims such as "Austin's best detailer" or customer ratings, because none were supplied
- A claim-check table linking statements like "we bring our own water and power" to the matching proof point
- A test plan pairing two angles and naming the one variable that differs, with no predicted click-through rate
This is an illustrative scenario showing how to fill in the prompt and what to look for. It is not a recorded model output. Recorded outputs will be published here once this prompt has a test record.
Testing and editorial review
Status: Untested. Reviewed for structure and clarity by the editorial team. No recorded model test yet.
No test record has been published for this prompt yet. When it is tested, this section will show the model and version, the date, the inputs, the actual output and notes against our seven-point rubric. We do not publish numerical scores.
| Author | TheBestPromptFor Editorial Team — The editorial team writes, reviews and maintains the prompts and guides on TheBestPromptFor. Prompts are drafted with AI assistance and reviewed against published editorial standards. |
|---|---|
| Published | November 15, 2026 |
| Last reviewed | November 15, 2026 |
| Last tested | Not yet tested |
| Version history |
|
Limitations and things to check
- Character counts from a model are not always exact. Confirm them in the ad platform.
- The [POLICY CHECK] flag is a prompt for you to review, not a ruling. Platform policies and advertising law are yours to verify.
- The model has no access to your account data, so it cannot know which messages have already been tried or how they performed.
- Copy is one part of an ad. Targeting, creative, bids and the landing page can outweigh any wording change.
Alternative prompts
Content Writing
Conversion Copywriting Prompt
Works out what a page must make the reader believe, then writes the copy in that order using only the proof you supply.
- ChatGPT
- Claude
- Gemini
Content Writing
Headline Writing Prompt
Generates headlines from several distinct angles, checks each against accuracy and length tests, and recommends three with reasons.
- ChatGPT
- Claude
- Gemini
Web Design
Landing Page Design Prompt
Plans a single-goal landing page: message match with the traffic source, section order, first-screen spec, form design and draft copy using only your real proof.
- ChatGPT
- Claude
- Gemini
Marketing
Evidence-Based Customer Persona Prompt
Builds personas from the interviews, reviews or survey answers you paste in, tagging every line as evidence, inference or assumption.
- ChatGPT
- Claude
- Gemini
Related resources
Technical guide
How to Test AI Prompts
A seven-criterion rubric and a nine-field test record for evaluating prompts, plus why this site describes results in words and publishes no scores.
Guide
How to Write a Good AI Prompt
The five parts of a complete prompt (goal, context, task, constraints, output format), with a worked rewrite and a pre-send checklist.
Frequently asked questions
Which platforms does this prompt work for?
Any text-based ad format. The structure stays the same and the format spec field carries the differences: paste the components and limits for the placement you are buying. For video placements it will write on-screen text and captions, but you will need a separate script for the footage itself.
Why does it ask me for character limits instead of knowing them?
Ad platforms revise their formats, and a model's recollection of them may be out of date. Pasting the current limits from the platform's documentation removes that uncertainty. If you skip the field, the prompt makes the model state its assumed limits so you can check them.
How many variations should I test at once?
That depends on your budget and traffic, since each variation needs enough impressions to tell you anything. The test plan starts with two angles for that reason. Add more only when the first comparison has run long enough for you to see a clear difference.
Sources and references
- Prompt engineering guide — OpenAI
- Prompt engineering overview — Anthropic
Was this prompt useful?
No account needed. Do not include personal information.