On this page
Start with the brief
The quality of the output depends on the input. Give ChatGPT what you sell, who it is for, the single conversion you want, your offer, the objections buyers raise, and the exact landing page text. Paste five to ten real phrases from reviews or sales calls, because customer language beats invented copy.
Without this it produces generic ads that match every competitor. With it, you get options that sound like your business.
Add constraints that matter: words you cannot use, claims you cannot make, a brand voice, and what competitors say. Saying what to avoid improves the result more than adding more detail about what you want.
Keep a reusable brief for each product or service. When you paste the same tested brief each time, the output is more consistent, and you can improve the brief when a result disappoints, instead of starting again from scratch.
Write responsive search ad assets
A responsive search ad takes up to 15 headlines of 30 characters and up to 4 descriptions of 90 characters. Ask for options in those limits, grouped by angle: price, speed, proof, guarantee, a feature. Then count the characters yourself, because language models miscount.
Use the output as raw material. Keep the lines that include the keyword and a clear claim, and cut the filler. See responsive search ads and Google Ads headline examples for what to aim for.
Test the output the way you would test your own copy. Run several distinct angles, let the account show which wins, and then ask for more in the style of the winner.
Pin headlines when you must show a specific line, such as a legal phrase or brand name, but use pinning sparingly. Too many pins reduce the combinations Google can test, which can lower the ad’s strength and its results.
Want this on a schedule?
Runs recurring checks on a schedule you set. Every change waits for your approval.
Use it on keywords and search terms
For keyword research, ask for themes and modifiers around your offer, then confirm volume and cost in a keyword tool. ChatGPT does not have your market’s search data and will invent plausible phrases.
The strongest use is on your own data. Export the search terms report, paste a sample, and ask it to group the terms by intent and flag ones that look irrelevant. You then decide which become negative keywords. The report itself is explained in the search terms report guide.
You can also ask it to explain a bad search term in plain language. “Why would someone type this, and what do they want?” is a quick way to decide whether a term is a customer or a researcher, which is the call that matters for negatives.
What not to hand over
Do not use it to set bids, budgets or targets from memory, or to say how your campaigns are performing unless you gave it the data and checked the answer. Do not let it make claims about your product you cannot support, which can lead to disapprovals. Check Google’s ad policies before launch.
Do not paste customer personal data or confidential figures into a chat you do not control. Anonymize first.
Check generated headlines for repetition. Language models tend to produce many lines that say the same thing in different words, and Google may show combinations that read oddly together. Make each line stand alone and say something distinct.
Prompts that work
For ad copy: “Here is my offer, audience and landing page text. Write 15 headlines under 30 characters and 4 descriptions under 90 characters for a Google responsive search ad. Group them by angle and use the customer phrases below.” Then paste the phrases.
For search terms: “Here are 100 search terms with cost and conversions. Group them by intent. List the ones that look irrelevant for a business that sells X, and explain why for each.” You review and decide.
For landing page match: “Here is an ad and the page it leads to. What does the page promise that the ad does not, and the reverse? Suggest changes to make them consistent.” This is a quick way to improve relevance. See landing pages for ads.
When a chat window is not enough
A chat session cannot see or edit your account. You copy results out, apply them by hand and repeat. For the repeated work, scripts help, see automate Google Ads.
Sprites connects to your Google Ads account, reads it, and proposes keyword, negative, ad and budget changes as approval cards you edit and approve before anything is applied.
Some teams use AI tools connected to the account through a standard protocol, so the assistant can read data directly. That removes the copy and paste, but the same rules apply: check the output, and keep changes reviewable. See what MCP is.
Questions
Can ChatGPT run my Google Ads for me?
Not on its own. In a chat window it cannot access your account. It can help write ads and analyze data you paste, but you apply and verify the changes.
Can ChatGPT write Google Ads copy?
Yes. Give it a detailed brief and ask for headlines of up to 30 characters and descriptions of up to 90. Check the counts, claims and policy before use.
Does Google Ads have its own AI features?
Yes. Google Ads includes its own generative tools for creating assets and suggestions inside the interface. Treat them as drafts, the same as any other AI output.
Is ChatGPT good for keyword research?
It is good for brainstorming themes, but it does not supply search volume or cost. Verify every idea in a keyword tool or in your own search terms data.