Ad creative doesn’t just need to look good, it needs to perform. Marketers are turning to ChatGPT to brainstorm, analyze, and improve ad copy faster. But without clear context, metrics, or structure, results are hit or miss.
If it doesn’t sell, it isn’t creative.
We analyzed the 150k conversations our own users have had with Prism (our AI marketing OS) to uncover what prompts worked best.
This guide represents the best of the best, and gives you 20 copy-and-paste prompts that are proven to work. You’ll learn exactly how to brief ChatGPT, what inputs to include, and why these techniques consistently outperform generic prompt lists.
The prompts below are a quick way to get yourself started on the basics of context engineering: the biggest skill gap most marketers run into when trying to use AI in their daily workflows.
Quick note:
At Pixis, we use Prism, which connects directly to your ad data across Meta, Google, TikTok, and more. Some of these ChatGPT prompts include instructions to upload a csv file or an image for the AI to analyze. With Prism, this step isn’t necessary. Just ask and get the answers you need.
Start a 14-day free trial today.
Quick Start: 5 ChatGPT Prompts for Writing High-Impact Ad Copy
Below, I’ll break down how to give ChatGPT the context it needs to do great analysis for you.
But before diving into full analysis prompts, here are five quick, ready-to-use prompts for writing ad copy. Copy and paste these directly into ChatGPT, swap in your brand and offer, and iterate.
1. Facebook/Meta Ad Headlines
“Act as a Meta Ads copywriter. Write 5 high-converting headlines for [product], targeting [audience]. Focus on [benefit] and include a call to action. Keep each under 8 words.”
2. Google Ads Headlines & Descriptions
“Act as a Google Ads specialist. Write 3 RSA headline sets (30 characters each) and 2 descriptions (90 characters) for [offer]. Use urgency and proof but no clickbait.”
3. TikTok Hook Lines
“Act as a short-form video strategist. Write 5 TikTok hooks for [product] targeting [audience]. Each should open with emotion or surprise, under 6 words.”
4. Retargeting Ad Variations
“You’re a performance marketing creative strategist. Write 3 ad versions for users who viewed [product] but didn’t convert. Include the offer and emotional reassurance.”
5. Ad CTA Copy Generator
“Write 10 concise CTAs (2–4 words each) for a campaign promoting [product/offer]. Each should imply urgency or benefit without sounding aggressive. Our audience is [brief audience description].”
Before You Start: Setting Up ChatGPT for Ad Analysis
Hopefully that gives you some easy, quick wins. Now let’s get a little more advanced. The prompts below are set up to give ChatGPT far more context specific to your task at hand.
So before running any of these prompts, you’ll need to prepare ChatGPT correctly.
This approach to prompt engineering is informed by thousands of Prism users and over 150k real marketing prompt sessions. The principles here reflect what we’ve seen consistently drive reliable and actionable outputs.
A Framework for Writing Successful Prompts
The most successful prompts share these traits, validated across Prism’s real user data:
- Role assignment: “Act as a performance marketing analyst.”
- Context clarity: Include platform, timeframe, audience, and goal.
- Specificity: Direct, measurable requests.
- Quantitative anchors: Define CTR, CPA, or ROAS thresholds.
- Set output expectations and formatting requests: Use tables, bullets, or summaries.
These rules are the foundation for every prompt in this guide, and why they consistently produce results.
Gather and Upload the Right Data
Pull raw reports from Meta Ads Manager or Google Ads in CSV or Excel format.
Include, at minimum:
- Ad ID or creative ID
- Headline, primary text, or hook
- Impressions, clicks, spend, conversions
- Date range and campaign name
If you’re evaluating visuals, upload actual ad creatives alongside your data for richer insight.
Always reference filenames and columns explicitly so ChatGPT understands what it’s analyzing.
For example:
“Here is a file: meta_perf_last30.csv.
It has the following columns: creative_id, headline, hook, CTR, CPA”
Define Metrics and Outcomes
Asking for “recommendations that improve performance” sets up ChatGPT to fail. It’s not specific enough.
Successful prompts name the exact KPIs ChatGPT should analyze (CTR, CPA, ROAS, MER, CVR, etc.) and define thresholds (e.g., “Top-performing = top 10% CTR, CPA under $20”).
These anchors give the model quantitative boundaries and prevent vague interpretations.
Outline the Steps and Expected Output
Every strong prompt provides an explicit sequence of steps you expect the LLM to take.
Even providing simple instructions like, “load data → calculate metrics → summarize trends” helps ChatGPT reason clearly and stay on task.
Also, tell ChatGPT exactly what format you want it to use in its response. Tell it to provide a table, charts, or bullets and that’s what you’ll get. Don’t ask it specifically, and it’ll choose on its own - often wrongly.
13 More Detailed ChatGPT Prompts for Ad Creative Analysis and Optimization
Each prompt below includes what it does, when to use it, required inputs, and the prompt itself in italics. Copy and paste it into ChatGPT, making sure you’ve got the accompanying files.
You may find you have to adjust how you format your files or describe the columns they contain. I cannot stress enough how worth it is to take the time to do that.
6. Competitor positioning angles and differentiation hooks
Use this prompt when you need to take stock of competitors’ messaging and identify whitespace.
Inputs: Competitor name and links to ad or landing pages.
For even better results, use ChatGPT’s agent mode and tell it to peruse Meta’s Ad Library or Google Transparency Center. It’ll take longer, but the added specific context goes a long way in helping the AI give you an answer you can use.
Prompt:
“Role: Act as a positioning strategist.
Inputs:
- Competitor: [competitor name]
- Sources: Ads Transparency/Meta Ads Library, [brand SERP], [landing pages]
Objective: Summarize competitor angles and propose differentiated hooks.
Steps:
1) Collect 10 ads and 3 landing pages.
2) Extract value props and group them into angles.
3) Identify whitespace opportunities.
Output:
- Table: angle, exemplar copy, frequency
- 3 new hooks (headline + support line)”
7. Persona-based pre-launch read on an ad
Use this prompt to test the potential resonance of copy before launch. ChatGPT is surprisingly good at taking on a persona and simulating audience reactions to ad copy, provided you give it a clear understanding of the persona it’s meant to adopt.
Inputs: A detailed persona description. You’ll also need an ad copy file (this example uses the example of a csv file of multiple ads, but you could just copy/paste in the ad copy for one ad).
Prompt:
Role: Act as a target persona reviewer.
Inputs:
- File: [ad_copy.csv]
- Columns: [Headline], [primary ad copy], [CTA]
- Persona: [age_range], [context], [pain point], [awareness stage], [demographics]
Objective: Simulate first-impression feedback.
Steps:
1) Read the ad.
2) Score the ad on clarity, relevance and emotion on a scale of 1 (least) to 10 (most).
3) Suggest one change to improve the resonance of the ad as much as possible.
Output:
- Scores + rationale table
- 3 bullets: what works, what confuses, what to test next
8. Variant scoring with anchored definitions and the psychology of advertising
Use this prompt when you have too many multiple copy variants to evaluate and need to gut check which is likely to perform best before you have real test data.
This prompt is intentionally agnostic of the audience, relying instead of first principles from the psychology of advertising that apply regardless of product or buyer persona. You could adapt it by adding a detailed description of your ICP for slightly different results, but we’ve found the template below works best.
Inputs: CSV with headline, primary ad copy.
Prompt:
Role: Act as a creative QA analyst.
Inputs:
- File: [advariants.csv]
- Columns: Headline, primary ad copy
Rubric Anchors:
- Clarity score 10 = entirely unambiguous and ≤12 words
- Emotion score 10 = evokes strong emotions
- Curiosity score 10 = extremely intriguing
Steps:
1) Use core principles of advertising psychology and the science of persuasion to evaluate each variant in the attached csv “[advariants.csv]” on a scale of 1-10 using the rubric to anchor high scores.
2) Provide justification for each score, referencing proven buyer psychology principles.
3) Suggest 1 improved variant per ad with evidence-based reasoning.
Output:
- Table: id, clarity score, emotion score, curiosity score, justification, recommended improvement.
9. New angles for a different audience segment
Use this prompt when your ad creative resonates only with a narrow set of your potential buyers, and you need to adapt the ads to work better for another segment.
Inputs: Audience-segmented creative data.
For this prompt, I’ve listed some columns that would be helpful to have, but make sure you customize this prompt carefully for your own use. The key is to give ChatGPT some success metric (ideally CTR), segmented by only one audience dimension (like age or geography), and some context about the ad creative and copy.
You might want to follow this prompt up with ChatGPT’s Agent tool turned on, and ask it to look at Meta Ads Library or Google’s Transparency Center for your competitors’ ads, and adjust its answer based on what it finds.
Prompt:
Role: Act as a growth copy strategist
Inputs:
- File: [winners.csv]
- Columns: theme, headline, offer, cta, creative, CTR, ROAS, [demographic segment - age, geo, income, etc.]
Objective: Extend themes to [target audience segment that needs improvement]
Steps:
1) Filter high CTR & ROAS for each age group, selecting only data with high confidence intervals.
2) Extract tone & visuals. Identify what resonates best with each age group, how it’s different from what resonates with audience members of other age groups, and explain why.
3) Create 5 new angles that would allow us to better reach [target audience segment].
Output:
- 5 new angle ideas (headline + visual cue)
- Summarize assumptions you’ve made about the target audience and justify your recommendations.
10. Alternative narrative frames for testing
Use this prompt when you want new narrative angles for creative testing.
Whatever ChatGPT gives you in response to this prompt, keep in mind that this is essentially an exercise in brainstorming. You shouldn’t necessarily go ahead and test its recommendations as-written.
Instead, use them as creative fuel that broadens the scope of your brainstorm, giving you more ideas, faster.
Inputs: Performance CSV with ad headlines, copy, CTR, impressions, and ROAS.
Prompt:
Role: Act as a narrative strategist.
Inputs:
- File: [creative_perf.csv]
- Columns: ad ID, headline, primary copy, spend, impressions, CTR, ROAS.
Steps:
1) Confirm the current best creative messaging angle as measured by CTR. Ignore ads with impressions < [impressions minimum]
2) Confirm the current winning angle as measured by ROAS. Ignore ads with spend < [spend minimum]
3) Cross-reference best performing ads for CTR and ROAS to uncover themes and successful narrative frames.
4) Generate 3 new narrative frames: [aspirational], [emotional], [authority].
5) Provide three headlines + supporting lines for each narrative frame.
Output:
- Bulleted list with sub-bullets
11. Executive-ready summary of creative performance
Use this prompt when summarizing campaign results for leadership.
This prompt is a quick-and-dirty way to produce an executive summary of performance across channels. You’ll need to automate exported reports from the ad platforms of your choice and re-attach them each time you want a dashboard with the most up-to-date data.
Inputs: Reports from the platforms you need to report on, in csv format. Be very careful to select matching date ranges on your exports. Also, if you export ad copy from one platform, it’s best to include it in all platforms, so as to avoid confusing the AI.
Prompt:
Role: Act as an executive communications editor.
Inputs:
- File: [Meta_campaign_perf.csv]
- File: [Google_campaign_perf.csv]
- File: [Shopify_sales_perf.csv]
Steps:
1) Analyze the results from each data source.
2) Identify performance highlights: three performance wins worth mentioning to my CMO, and three areas for improvement.
3) Summarize performance of each channel independently.
4) Identify cross-platform trends and performance themes in creative, demographics, product, and offer.
5) Recommend prioritized next steps to maximize [target KPI].
Output:
- 3 bullets under 120 words each.
12. Generate a Creative Brief from Last Month’s Top Ads
Use this prompt to turn performance data into a clear, ready-to-use creative brief. It helps distill what worked, why, and how to replicate that success in your next campaign.
Inputs: A CSV of last month’s top-performing ads.
Prompt:
Role: Act as a creative director.
Inputs:
- File: [last_month.csv]
- Columns: headline, primary text, CTR, ROAS, spend
Steps:
- Filter ads where CTR ≥ [3%] and ROAS ≥ [4].
- Identify shared themes, tones, and audience traits.
- Extract key messages and visual cues.
- Summarize findings into a brief that a copywriter or designer can act on.
Output:
Sections: Objective, Target Audience, Tone, Key Messages, Example Headline.
13. Create a One-Slide Summary of Google Ads Performance
Use this prompt to compress a dense Google Ads report into a single, presentation-ready summary. Ideal for fast updates or stakeholder decks.
It outputs text, not an actual slide, but the important thing here is the analysis included and the way it’s presented.
Inputs: Google Ads export CSV.
Prompt:
Role: Act as a marketing reporting specialist.
Inputs:
- File: [google_ads.csv]
- Columns: campaign, CTR, CPC, CVR, spend
Steps:
- Calculate week-over-week changes for CTR, CPC, and CVR.
- Identify 3 notable creative or keyword insights.
- Summarize top takeaways for decision-makers.
Output:
Title + 3 bullets (≤14 words each) + one-sentence caption.
14. Summarize Themes Driving Growth and Fatigue
Use this prompt to understand what’s scaling and what’s losing steam across your creative library. It’s especially useful for identifying when once-strong concepts start to decline.
We’ve seen customers use this as a way to prepare themselves for meetings between the creative and performance teams, or just to help find new ideas in a brainstorming session.
Inputs: Meta performance export with creative themes.
Prompt:
Role: Act as a campaign storyteller.
Inputs:
- File: [meta_export.csv]
- Columns: theme, CTR, impressions, spend, ROAS
Steps:
- Identify top-volume creative themes.
- Detect declining CTR or ROAS trends.
- Write 3 short paragraphs explaining:
- What’s driving volume
- What’s showing fatigue
- What actions to take next
Output:
3 concise paragraphs (drivers, fatigue, next steps).
15. Weekly Creative Check-In and Next Steps
Use this prompt for quick weekly reviews. It provides a snapshot of what’s performing, what’s fading, and what to test next.
This one is most useful when you start to automate the underlying report you’re uploading from Google Ads, Meta Ads, or something else.
Do that, then drop the weekly report csv into this prompt and you’re off to the races.
Inputs: 7-day performance export.
Prompt:
Role: Act as an ads optimization analyst preparing for a weekly test sprint meeting.
Inputs:
- File: [last7days.csv]
- Columns: creative_id, CTR, ROAS, spend
Steps:
- Rank creatives by performance.
- Flag signs of fatigue (declining CTR or ROAS).
- Recommend the next three test priorities.
Output:
3 bullets: Top Performer, Fatigue Watch, Next Test.
16. Build a Three-Step Creative Testing Roadmap
Use this prompt to plan your next testing sprint. It structures priorities into clear “Pause,” “Scale,” and “Create” actions, so your test plan is always tied to data.
Be mindful here: ChatGPT can only simulate what it thinks it would be like to be an expert in running advertising experiments. It’s not an actual expert.
But because it’s not an expert, it sometimes is better at seeing the most obvious areas of improvement. Use this prompt to get responses from ChatGPT that prompt your own creativity - don’t rely on its replies as-given.
Inputs: Performance CSV with CTR, CPA, ROAS, and spend.
Prompt:
Role: Act as an ad test planner.
Inputs:
- File: [perfromance.csv]
- Columns: ad_id, CTR, CPA, ROAS, spend
Steps:
- Pause ads with ROAS below X or CPA above Y.
- Scale ads with ROAS above A and CTR above B.
- Create 3 new test hypotheses based on what’s working.
Output:
Table: Action | Criteria | Rationale.
17. Audit Brand Consistency Across Ads
Use this prompt to verify whether your ad copy aligns with your brand’s voice and message. It’s especially valuable before scaling spend or localizing creative.
ChatGPT is surprisingly good at this one on its own. Every LLM works by essentially matching how similar different bits of text are to one another. So when you have an ad that talks about “value” while your brand guidelines focus on “exclusivity”, it stands out to the AI.
Even so, as with other prompts in this list, you shouldn’t consider running this prompt a complete, verified check. But it is a fast way to find the most obvious offenders violating brand guidelines (if there are any).
Inputs: Ad copy in a csv and brand guidelines.
Prompt:
Role: Act as a brand editor.
Inputs:
- Ads: [ads.csv]
- Brand Guide: [brand_guide.md]
Steps:
- Compare tone, phrasing, and claims against the brand guide.
- Flag misalignments or inconsistent messages.
- Recommend specific, minimal edits that fix the issue.
Output:
Table: ad_id | Issue | Excerpt | Suggested Fix.
18. Unify Brand Promises into a Single Narrative
Use this prompt when your ads are saying too many different things. It helps consolidate scattered value propositions into one coherent brand story.
As it’s written, this prompt is built to evaluate all your creative. You might also try this one with separate csv files, each containing the ad creative you’re serving to different demographics, geographies, or products.
Inputs: CSV of ad headlines with CTR and impressions.
Prompt:
Role: Act as a brand strategist.
Inputs:
- File: [headlines.csv]
- Columns: headline, CTR, impressions
Steps:
- Cluster repeating promises or claims.
- Rank each cluster by CTR to find what resonates most.
- Write one unified brand narrative with 3 proof points drawn from the data.
Output:
Short narrative paragraph + 3 supporting bullets.
Prompt Engineering Principles for Marketers
To get reliable, repeatable results:
- Define metrics explicitly. Always name the KPIs and thresholds (CTR, CPA, ROAS, etc.).
- Assign a role. Guide ChatGPT’s tone and reasoning process by defining who it should “be.”
- Add structure. Request tables, bullets, or paragraphs to control formatting.
- Clarify subjectivity. Define terms like “top,” “winning,” or “high-performing.” Better yet, avoid them altogether, and use your specific metric-based definition, like “ads with a higher than 4% CTR”.
- Iterate. Ask ChatGPT to refine or re-rank based on prior outputs.
Download:
- CSV templates for Meta and Google Ads with required columns.
- Prompt checklist: Role → File (columns) → KPIs → Steps → Output format.
Conclusion:
By definition, an AI cannot be creative. It’s a machine that turns text into numbers, then looks for similarities between those numbers and predicts the “next best text” to reply to you.
But that doesn’t mean AI can’t help you act more creatively, and do it faster.
In fact, most of the reason marketers sometimes report AI underperformance, it’s simply because tey haven’t provided their LLM of choice with enough context. Combined with real data, these prompts help you test faster, learn smarter, and consistently drive lift.
Want to see how smart AI can be when it’s trained on 3B performance marketing data points from thousands of ad campaigns?
Give Prism a try, free for 14 days.