How to use ChatGPT to write Google Ads copy that performs (with prompts)
Learn a 4-step workflow for using ChatGPT to write responsive search ad copy that meets Google Ads character limits, support...
The problem with responsive search ads is volume. Google recommends using all 15 headline slots and all 4 description slots to give its machine learning enough combinations to test. For a single campaign with multiple ad groups, that means dozens of short, distinct, non-repetitive assets. Writing that manually takes hours. Most teams end up with thin, repetitive copy by the end.
The average Google Ads manager spends 40% of their time on tasks AI can automate, and most of that time goes toward copywriting and variation work. This guide walks through a 4-step workflow for using ChatGPT to write responsive search ad copy that holds up under real campaign conditions, with prompts you can copy and adapt today.
Why RSAs make ad copywriting harder than it used to be
Responsive search ads replaced expanded text ads in June 2022. Since then, Google selects which headlines and descriptions appear together, not the advertiser. Every asset must work on its own and pair well with any other asset in the ad. One RSA requires up to 15 headlines and 4 descriptions, each written to a strict character limit. That writing task adds up fast.
Ready to grow your organic traffic?
Get a free SEO audit from the Launchcodex team.
Before RSAs, writing three to five ad variations was enough. Now Google's machine learning tests combinations automatically. That works well when inputs are strong and breaks down when they are not. A poorly varied set of headlines restricts what the algorithm can test and lowers Ad Strength, Google's indicator of how well an ad is set up for performance.
RSA character limits at a glance
| Asset type | Character limit | Max number of assets |
|---|---|---|
| Headline | 30 characters | 15 |
| Description | 90 characters | 4 |
| Display URL path | 15 characters each | 2 paths |
Google recommends using all 15 headline slots. Each headline must stand alone because Google may display any combination of three headlines alongside any two descriptions. Fifteen inputs create up to 900 possible headline combinations. Volume without variety gives the algorithm nothing useful to learn from.
Why most teams end up with thin copy
Writing 15 distinct, non-repetitive headlines in 30 characters each is hard. Teams run out of angles after headline 8 and start repeating the same phrases with minor word swaps. The result is low Ad Strength, limited testing signal, and copy that sounds like it was drafted by committee.
ChatGPT solves this specific problem well. As the team at Position2 noted after a hands-on experiment: the challenge is to present the same information "in enough different ways to populate the maximum number of allowable ad copy variants for Responsive Search Ads. This is what ChatGPT does really well."
What to feed ChatGPT before you write a single prompt
The quality of ChatGPT's output depends entirely on the quality of the input. A vague prompt produces generic ads. A structured brief produces campaign-ready copy. Before writing any prompt, gather the six inputs below and feed them to ChatGPT at the start of every session. This is the step most guides skip, and it is the one that changes everything.
Google Ads consultant Zdravko Dimchov puts it plainly: "Bad prompts equal generic ads. Good prompts equal campaign-ready outputs." Most teams skip the brief and jump straight to asking ChatGPT for headlines. That is why the output sounds like every other ad on the page.
"Most people treat ChatGPT like a search engine. They ask a vague question and expect a useful answer. Ad copy prompts work exactly the same way as any other AI workflow. The better your inputs, the less time you spend fixing outputs."
Derick Do, Co-Founder & Chief Product Officer, Launchcodex
The six inputs ChatGPT needs
- Product or service description. One to three sentences describing exactly what you sell, the primary benefit, and who it is for.
- Target audience. Who is searching. Include pain points, intent signals, and what they are trying to accomplish.
- Primary keyword. The main search term the ad group targets.
- USPs. Three to five specific differentiators. Skip vague phrases like "quality service" or "great results."
- Tone. Direct and professional, conversational, urgent, or whichever tone fits the brand.
- Call to action. The exact action you want the reader to take: get a quote, start a free trial, book a call, shop now.
Example brief for a B2B SaaS product
Product: Project management software for remote engineering teams. Tracks tasks, sprints, and blockers in one workspace. Integrates with GitHub and Slack.
Audience: Engineering managers at companies with 20 to 200 employees. Frustrated with scattered task tracking across Jira, Slack, and spreadsheets. Want one place developers actually use.
Primary keyword: project management software for engineering teams
USPs: GitHub integration in two clicks, built for developers not project managers, used by 4,000 teams, 14-day free trial with no credit card required.
Tone: Direct and confident, slightly technical.
CTA: Start free trial.
Feed that brief to ChatGPT before asking it to write anything. It gives the model the same context a human copywriter would need before sitting down to draft.
How to structure prompts that actually work
Effective Google Ads prompts follow four components: context, objective, constraints, and output format. Missing any one of these produces output that needs heavy editing or cannot be used at all. The constraints section, where you specify character limits and asset counts, is the most commonly skipped and the most critical.
ChatGPT does not know Google Ads rules unless you tell it. Left without constraints, it writes headlines that run 45 to 50 characters long and descriptions that spill past 90. Those outputs cannot go live without complete rewrites.
The four-part prompt structure
Context. Tell ChatGPT who it is. Role prompting anchors the output style and expertise level.
"Act as a senior Google Ads copywriter with 10 years of experience writing high-converting responsive search ads."
Objective. Tell it exactly what you want. Be specific about the campaign goal.
"Write responsive search ad copy for [product] targeting [keyword]."
Constraints. State every limit explicitly and ask ChatGPT to verify compliance before including each asset.
"Write 15 headlines. Each headline must be 30 characters or fewer, including spaces. Count the characters before including each headline. Rewrite any that exceed 30 characters. Write 4 descriptions. Each must be 90 characters or fewer."
Output format. Tell it how to return the output so you can use it without reformatting.
"Format the output as a numbered list. Label headlines H1 through H15 and descriptions D1 through D4. Show character count in parentheses next to each asset."
Weak prompt versus strong prompt: a direct comparison
Weak prompt:
"Write Google Ads headlines for a project management app."
ChatGPT returns headlines that may run 40 to 55 characters, repeat the same value proposition four ways, and include the brand name where it does not fit.
Strong prompt:
"Act as a senior Google Ads copywriter. I am writing responsive search ads for a project management app for remote engineering teams. Primary keyword: project management software for engineering teams. USPs: GitHub integration, built for developers, 4,000 teams use it, 14-day free trial.
Write 15 headlines. Each must be 30 characters or fewer including spaces. Count every character before listing each one. Write 4 descriptions, each 90 characters or fewer. Tone: direct and confident. CTA: start free trial. Format as a numbered list with character count in parentheses."
The strong prompt produces usable, on-spec output. It still needs an editing pass, but the gap between raw output and uploadable copy is measured in minutes, not hours.
The prompts: copy-ready templates for RSA headlines and descriptions
These four prompts cover the most common RSA copy tasks. Each includes the constraint language needed to enforce character limits inside ChatGPT. Copy each one, fill in the brackets, and run it with your brief loaded at the start of the session. Do not skip the brief step.
Prompt 1: full RSA copy from a brief
Use this when starting a new ad group from scratch.
Act as a senior Google Ads copywriter with deep knowledge of responsive search ad best practices.
Here is the brief:
- Product/service: [describe in 2-3 sentences]
- Target audience: [describe pain points and intent]
- Primary keyword: [keyword]
- USPs: [list 3-5 specific differentiators]
- Tone: [direct / conversational / urgent / other]
- CTA: [action]
Write 15 headlines. Each headline must be 30 characters or fewer, including spaces.
Count every character before including a headline. If it is over 30 characters, rewrite it
until it fits. Do not submit any headline over the limit.
Write 4 descriptions. Each must be 90 characters or fewer, including spaces.
Format output as a numbered list. Label headlines H1-H15 and descriptions D1-D4.
Show character count in parentheses next to each asset.Prompt 2: improve existing copy
Use this when you have ads running and want to lift performance without starting from scratch.
Act as a senior Google Ads copywriter and conversion rate specialist.
Here is my current responsive search ad copy:
[paste existing headlines and descriptions]
Brief context:
- Product/service: [describe]
- Primary keyword: [keyword]
- Current performance note: [CTR, conversion rate, or Ad Strength rating if known]
Review this copy. Identify:
1. Headlines that are too generic or not benefit-driven
2. Descriptions that are vague or do not include a CTA
3. Missing angles such as urgency, social proof, or specificity
Rewrite the weak assets. Keep strong ones. Every rewritten headline must be 30 characters
or fewer. Every rewritten description must be 90 characters or fewer. Show character counts.Prompt 3: generate thematic variations for split testing
Use this when you want two or three complete RSAs with distinct messaging angles to test against each other.
Act as a senior Google Ads strategist and copywriter.
Brief:
- Product/service: [describe]
- Keyword: [keyword]
- USPs: [list]
- CTA: [action]
Create 3 complete responsive search ads. Each ad must have its own clear messaging theme:
- Theme 1: Focus on [benefit 1, e.g., speed or time savings]
- Theme 2: Focus on [benefit 2, e.g., social proof or results]
- Theme 3: Focus on [benefit 3, e.g., ease of use or simplicity]
For each theme, write 5 unique headlines (30 characters or fewer each) and 2 unique
descriptions (90 characters or fewer each). Show character counts. Each theme must feel
distinct so Google can test them as separate strategic angles, not just word variations.Prompt 4: differentiate from competitor ads
Use this when you want copy that stands apart from what competitors are running on the same keyword.
Act as a competitive Google Ads copywriter.
Here are competitor ads running for [keyword]:
[paste competitor headlines and descriptions from Google search results]
My brief:
- Product/service: [describe]
- USPs: [list]
- CTA: [action]
Analyze these competitor ads. Identify:
1. Messaging they all share (what to avoid repeating)
2. Angles no competitor is using (opportunities)
Write 10 headlines and 3 descriptions for my ad that are clearly differentiated.
Every headline must be 30 characters or fewer. Every description must be 90 characters
or fewer. Show character counts.How thematic variation connects to split testing
RSA split testing works best when separate ads in the same ad group carry distinct messaging themes, not random headline swaps. When two ads argue different cases for the same product, the results tell you which value proposition drives clicks and conversions. ChatGPT generates multiple themed batches quickly once the brief is set.
Most teams treat RSA variation as a creative exercise. The goal is a structured test. When two ads in the same ad group share the same underlying message with slightly different words, there is nothing meaningful to learn from the results. Real theme contrasts produce real learning.
"We run this workflow on client campaigns regularly. The time win is real, but the bigger shift is copy quality. Themed RSAs give us something to actually learn from during testing, not just more volume for the algorithm."
Tanner Medina, Co-Founder & Chief Growth Officer, Launchcodex
How to set up a theme-based split test
- Identify two or three distinct value propositions for the product. Examples: speed, cost, trust, simplicity, results.
- Assign one theme to each RSA in the ad group.
- Use Prompt 3 above to generate each themed ad separately.
- Run the ads simultaneously for at least four weeks before drawing conclusions.
- Review which theme produces higher CTR and conversion rate.
- Remove the low performer, create a new variation of the winner with a different angle, and repeat.
A common mistake to avoid
The most common error is building ads that are thematically identical despite different wording. If one ad opens with "Grow your business faster" and another opens with "Scale your team with ease," those are not distinct themes. Both are vague benefit statements. A real theme contrast looks like this: "14-day free trial, no card" (proof of low risk) versus "4,000 engineering teams trust it" (social proof). One argument is about risk reduction. The other is about crowd validation. Those are testable differences.
How to edit ChatGPT output before it goes live
ChatGPT output is a starting point, not a finished product. Even with a well-structured prompt, some headlines will need character corrections, some will be too generic, and some will clash with brand voice. A short editing pass before upload is faster than starting from scratch and is required for any live campaign.
The team at Position2 ran a hands-on experiment with ChatGPT-generated Google Ads copy and found they could bring raw output "up to an acceptable standard in less than 10 minutes." That is the realistic benchmark. ChatGPT accelerates the draft. The strategist finishes it.
Pre-upload editing checklist
Run through these checks before uploading any ChatGPT-generated RSA copy.
- Count every headline character manually or paste assets into a character counter. ChatGPT miscounts occasionally.
- Remove any headline that repeats an exact phrase from another headline in the same ad.
- Confirm at least three headlines include the primary keyword or a close variant.
- Verify every description includes a call to action.
- Read each headline aloud. If it sounds awkward or forced, rewrite it.
- Flag any superlative claims like "best" or "number one" that lack supporting evidence. These can trigger policy disapprovals.
- Check that no headline contains the brand name unless it was intentional and fits within the 30-character limit.
- Run the final set through the Google Ad Preview and Diagnosis tool. More than 60% of Google searches happen on mobile, where wider characters can cause visual truncation even within the character limit.
A note on ChatGPT and live campaign data
ChatGPT does not connect to your Google Ads account by default. It cannot see your campaign metrics, search term reports, or historical ad performance. Every prompt in this article works with information you supply. If you want ChatGPT to analyze actual performance data, export from Google Ads and paste the data directly into the prompt before asking for recommendations.
How copy quality affects Ad Strength and Quality Score
Ad Strength and Quality Score are not vanity metrics. Both influence how often your ads show and what you pay per click. Copy decisions made during the writing phase directly affect both scores. Higher Ad Strength means more combinations tested, better auction efficiency, and lower cost per result over time.
Ad Strength rates RSAs on a scale from Incomplete to Excellent. Google calculates it based on how many assets you provide, how unique they are, and how well they reflect the ad's keywords and landing page. A set of 15 varied, keyword-aligned, benefit-focused headlines rates higher than 15 variations of the same sentence.
How ChatGPT-generated copy affects each Ad Strength factor
- Quantity. ChatGPT produces all 15 headlines and all 4 descriptions in one session. That removes the most common reason for low Ad Strength: teams stopping at 7 or 8 headlines because they ran out of ideas.
- Uniqueness. Thematic prompting ensures each headline carries a distinct angle. This directly improves the uniqueness component of Ad Strength scoring.
- Keyword relevance. Including the primary keyword and close variants in the brief ensures ChatGPT incorporates them naturally rather than forcing them in at the end.
Quality Score is calculated per keyword and reflects expected CTR, ad relevance, and landing page experience. Copy that closely matches what the searcher wants and delivers on the landing page lifts the relevance component. A higher Quality Score lowers CPC and improves ad rank without requiring higher bids.
AI-optimized creatives have shown the potential to deliver up to two times higher CTR compared to manually designed versions, and some campaigns have reported up to a 50% lift in return on ad spend after adopting AI-generated creative. Those outcomes depend on brief quality, prompt structure, and the editing pass that follows generation.
Three copy mistakes that hurt Ad Strength
- Over-pinning assets. Locking too many headlines to fixed positions removes the combinations Google can test. Agencies recommend pinning only critical elements like a required legal disclaimer or a specific time-sensitive offer.
- Repeating the same value proposition across all 15 headlines. Thematic prompts are designed to prevent this, but always check the output before uploading.
- Writing descriptions without a CTA. Descriptions should extend the headline's promise and tell the reader what to do next. Descriptions that only describe the product without directing action waste 90 characters.
Put the workflow into practice
ChatGPT is a real time saver for Google Ads copy when used with a proper brief, structured prompts, and a human editing pass before anything goes live. The 4-step process in this article, brief, prompt, generate, and edit, removes most of the friction from RSA copy production without removing the strategic thinking that makes ads convert.
Marketers using AI report saving more than 5 hours per week, and those using AI-generated content are 25% more likely to report campaign success. ChatGPT does not replace the strategist who knows the product, the audience, and what the data is saying. It accelerates the parts of the job that used to drain time before the real work started.
At Launchcodex, the paid media workflow our team follows is built on the brief-first principle. The better the input, the faster a campaign reaches its first meaningful test. If you want a team that already runs this workflow at scale, our performance media practice is a good place to start.
FAQ
Does ChatGPT stay within Google Ads character limits automatically?
No. ChatGPT frequently exceeds the 30-character headline limit unless you state the limit explicitly in the prompt and ask it to count characters before including each asset. The prompts in this article include that constraint language. Always verify character counts manually before uploading to your account.
Can I use ChatGPT to improve ads that are already live?
Yes. Prompt 2 in this article covers that workflow. Paste your existing headlines and descriptions, provide the brief context, and ask ChatGPT to identify weak assets and rewrite them. Including current performance data such as CTR or Ad Strength rating gives the model more to work with.
Does AI-generated copy violate Google Ads policies?
No, as long as the content itself meets Google's advertising policies. Google does not penalize ads based on how they were written. The standard rules still apply: no false claims, no misleading offers, no unsupported superlatives. Review all AI-generated copy against Google's ad policies before uploading.
How many ad themes should I test per ad group?
Two to three themes per ad group is the practical range. One theme gives Google nothing to compare. More than three makes it harder to interpret results clearly. Create one RSA per theme, run them simultaneously, and wait at least four weeks before reviewing which theme is winning on CTR and conversion rate.
What version of ChatGPT works best for Google Ads copy?
GPT-4 or later produces more reliable output for constrained tasks like RSA copy writing. It follows multi-step instructions more consistently than earlier versions and is less likely to exceed character limits. GPT-3.5 can work for basic copy generation but requires more editing passes to get on-spec output.
