Google Ads AI: write better ads and stop paying for junk searches

How can Google Ads AI help a small business?

The most useful Google Ads AI help for a small business is your own AI assistant, used for three specific jobs. First, it writes responsive search ad headlines and descriptions that fit Google's character limits. Second, it reads your search terms report and identifies the searches that cost money without ever producing a sale. Third, it turns those searches into negative keyword lists, so you stop paying for clicks that were never going to convert.

Google Ads also has its own built-in automation, which changes often. This guide concentrates on the writing, sorting and checking you do yourself with ChatGPT, Claude or Gemini, while you keep the final decision about the account.

You need three things to start: your product facts, a recent search terms report exported from Google Ads, and the web address of the page each ad sends people to.

What are the limits for responsive search ads?

A responsive search ad is the standard text ad in Google Search campaigns. You supply a set of headlines and descriptions, and Google combines them in different ways, showing whichever combinations it predicts will perform best for each search.

Your AI needs to know the limits before it writes a word, and as of November 2026, Google Ads Help on responsive search ads gives these figures.

Two more details from the same page are worth knowing. Pinning means fixing a headline to a particular position, such as always first. Google says pinning reduces the number of combinations it can build, so it isn't recommended for most advertisers. In languages such as Japanese, Korean and Chinese, each character also counts as two.

How do you write Google ads with AI?

Effective search ads repeat what the person searched for, give one solid reason to choose you, and say clearly what to do next. Your AI can draft all of that quickly, provided it works from real facts. Follow these steps for each ad group, meaning a set of closely related keywords that share the same ads.

Step 1: Give your AI the facts, not adjectives

Write a short fact list covering what you sell, your prices if you show them, delivery terms, guarantees, what makes you different and any proof you can back up. "Free returns within 30 days" is a fact, whereas "premium quality" is only an adjective. Ads built on facts are easier to believe, and you can defend every word if anyone questions it.

Step 2: Give it the keywords for this one ad group

Each ad group should cover a single idea, such as "running socks" or "emergency plumber". Paste its keywords into the chat, so your AI can write headlines that echo those words and make the ad feel relevant to the search.

Step 3: Ask for a spread of headline types

Fifteen headlines that express the same idea in different words give Google nothing meaningful to test. Ask instead for a deliberate mix that covers the keyword, your main benefit, specific proof, the offer and a clear call to action.

Step 4: Read every line as a buyer would

Remove capitals used for shouting, punctuation used for effect and words repeated only to fill space. Then read each headline alongside two others chosen at random, because Google can show any of them together, so every headline must make sense in any combination.

You are a Google Ads copywriter. Write assets for one responsive search ad.

Ask me for these first, if I haven't given them:
- The product or service, and the web page the ad links to
- The keywords in this ad group
- My facts: prices, delivery, guarantees, proof, what makes us different
- Words I must never use, and claims I can't make

Then write:
- 15 headlines, each 30 characters or fewer (spaces count)
  - 4 that include the main keyword
  - 4 about the main benefit
  - 3 with proof or a specific fact
  - 2 about the offer or terms
  - 2 calls to action
- 4 descriptions, each 90 characters or fewer
- 2 display path words, each 15 characters or fewer

Rules:
- Use only facts I gave you. If a headline needs a fact you don't have, leave it out and tell me.
- No exclamation marks, no ALL CAPS, no repeated words across headlines just to fill space.
- Each headline must make sense on its own and next to any other headline.
- Don't pin anything unless I ask.

Show a table with: asset, text, character count, type, and the fact it rests on.
Count characters carefully, including spaces. Mark any line over the limit as OVER.

How do you find wasted spend in the search terms report?

Your keywords are the terms you bid on, while search terms are what people actually typed, and the two are often quite different. The search terms report shows the real searches that triggered your ads and how each one performed. As of November 2026, you find it under Campaigns, then Insights and reports, then Search terms. Google omits some very low-volume searches to protect privacy, so the report never shows every search.

Step 1: Export enough data

Choose a date range long enough to be fair, such as the last 30 or 90 days, because a short window gives too little data to judge most searches fairly. Export these columns: search term, campaign, ad group, impressions, clicks, cost, conversions and conversion value. A conversion is whatever action you count as a success, such as a sale or a completed lead form.

Step 2: Know your break-even cost

Before your AI can label a search "wasted", it needs a threshold to measure against. Suppose your average order earns you $30 profit before advertising costs (an example figure). In that case, a search term that has cost $60 without a sale is a strong candidate to cut, whereas one that has cost $4 is too early to judge.

Step 3: Let your AI sort the terms

Ask it to place each search into one of four groups: winners, clear junk, judgement calls and too early to tell. Junk usually means the searcher wanted something you don't sell. Typical signals include words like "free", "jobs", "how to" and "DIY", a different product, or a location you don't serve. However, "how to" might be perfectly relevant if you sell a course, which is why your AI should explain each decision and you should make the final call.

Step 4: Look for repeated words, not only single searches

Many searches appear only once or twice, so individually they look harmless, but grouped by a shared word they can add up to real money. Ask your AI to group terms by repeated words and phrases (sometimes called n-grams) and total the cost of each group. Our free search term n-gram spend analyzer does this from a CSV export, entirely in your browser.

You are a careful Google Ads analyst. I am attaching a search terms report export.

My business: [what you sell, and what you don't sell]
Where I sell: [countries, cities or "online only"]
My break-even cost per sale: [for example, $30]
Date range of the report: [for example, last 90 days]

Do this:
1. Check the file. Tell me the columns you found, the number of rows, and the total cost. If cost or conversions are missing, stop and tell me.
2. Sort every search term into one group:
   - Winner: has sales at a cost per sale below my break-even
   - Junk: the searcher clearly wanted something I don't sell
   - Judgement call: could be junk or could be a buyer
   - Too early: spent less than half my break-even, no sale yet
3. Group the terms by repeated words or phrases. For each word, total the clicks, cost and conversions.
4. Draft negative keywords for the junk group. For each one, give the match type (broad, phrase or exact), the level (campaign or account list), and the search terms it would block.
5. List winners I could add as keywords.

Rules:
- Use only numbers from the file. Show your sums.
- Never suggest a negative that would block a search with a sale in this data.
- Explain each junk call in under 12 words.
- These are proposals for me to review. Don't tell me to apply them without checking.

Output: four tables (winners, junk with proposed negatives, judgement calls, word groups), then a short summary of the total cost in the junk group.

How do you build negative keyword lists?

A negative keyword tells Google not to show your ad for searches that contain it. Google's own illustration is an optometrist who sells glasses adding "wine glasses" as a negative.

Negative keywords behave differently from ordinary keywords, which is easy to miss. Google's negative keywords help page says they don't match close variants, so a negative for "free" won't block a misspelling or a plural you didn't add yourself.

The examples in the table come from Google's help page. Here is a simple way to organise your lists, from the broadest level to the narrowest.

• An account-wide list holds words that are never right for your business, such as "jobs", "careers", "free" or "second hand", provided those words never lead to sales.
• Campaign lists hold words that are wrong in one campaign but acceptable in another. For example, block your brand name in a non-brand campaign, so you can see each campaign's real results.
• Ad group negatives steer a search towards the right ad group, so that the most relevant ad appears for it.

Don't overdo it, because Google warns that too many negatives can reduce the number of customers you reach. Add negatives based on real search data, rather than copying an enormous list you found online.

• Every negative comes from a real search in your report, or from a clear rule about what you don't sell.
• No negative blocks a search that has already led to a sale in your data.
• Plurals, misspellings and alternative forms are added wherever they are needed.
• Each negative sits at the appropriate level, whether account, campaign or ad group.
• You have recorded the date you added each negative, so you can review it later.

Why does the landing page need to match the ad?

Someone searches, sees your ad, clicks and arrives on a page, and if that page doesn't answer their search, they leave after you have already paid for the click. Google also measures this. Its Quality Score has three components, one of which is landing page experience, meaning how relevant and useful your page is to people who click your ad.

Run this quick match check for each ad group before you spend more on it.

• The main words of the search appear in the page's first heading or on its first screen.
• The offer in the ad appears on the page with the same price, delivery terms and guarantee.
• The ad leads to the specific product it mentions, rather than the home page or a broad category.
• The next step the ad promises, whether buying, booking or requesting a quote, is easy to find on a phone.
• The page loads quickly on a mobile data connection, not only on office broadband.

For a complete review of the page itself, work through our landing page checklist.

A worked example: Bombas running socks

Bombas sells socks, underwear, t-shirts and slippers. Its giving model is "One Purchased = One Donated", meaning that for each item bought, it donates an item to people facing homelessness through partner organisations. It also has a running socks collection for men and women.

This example was written for this article, so these are not Bombas's real ads, keywords or results, and every number below is an example figure.

Writing the ad

The inputs are the running socks page, the ad group keywords ("running socks", "socks for running", "running ankle socks") and the fact list drawn from the brand's own pages. Here is part of an example draft that prompt 1 might return, with character counts.

Notice what is missing from the draft: there is no "best socks ever" and no claim about preventing blisters, because nothing in the fact list proves either one. A claim that isn't on the brand's own pages doesn't belong in the ad.

Sorting the search terms

Suppose the report contains a few hundred running sock searches. Prompt 2 might flag terms like "free running socks", "how to wash running socks" and "running socks jobs" as junk or judgement calls. It might also flag searches naming other sock brands as judgement calls, because whether to bid on competitors is a business decision rather than a data question. Searches like "running socks for women" that produced sales go into the winners group, and might deserve their own ad group and ad.

The result is a short list of proposed negatives with match types, each linked to the searches it would block. A person then reviews the list, removes anything risky and adds the rest to the account.

Common mistakes, and what to do instead

Several of these mistakes come from trusting the AI's output without checking it, so the fixes in the table are mainly about verification.

Where AI Blocks helps

These free items each handle one part of this job, and you copy the prompt or download the skill to use in your own AI. None of them change your account, because they only produce proposals for you to review.

• Ad headline writer produces plain headlines from your buyer and facts, with character counts and the single line to test first.
• Search term sorter takes your search terms reports and returns winners to promote, negatives to add and a waste score.
• Google Ads search-term reviewer gives you promising terms, suggested exclusions, judgement calls and the calculations behind them, measured against your profit target.
• Search query analyst specialist is a ready-made AI specialist that finds wasted spend and drafts negative keywords by level.
• Google Ads report auditor checks an export for wasted spend and shows the rows behind each proposed change.
• Non-brand Search campaign builder creates themed ad groups, keywords and ad copy from your products and search terms.
• Search campaign ad and page planner is a workflow that turns profitable searches into campaign drafts and matching page plans.

Search terms reports are spreadsheets, so our guide to analysing a spreadsheet with AI is useful here too. To keep your AI from inventing claims in your ads, read how to stop AI making things up.

Questions people ask

Can ChatGPT write Google ads?

Yes, ChatGPT, Claude and Gemini can all draft headlines and descriptions. Give them your facts, your keywords and the limits, which as of November 2026 are 30 characters per headline and 90 per description, and then check the counts and claims yourself.

How many headlines should a responsive search ad have?

Google requires at least 3 headlines and allows up to 15. Providing more gives Google additional combinations to try, although they only help when each one says something different.

How often should I check my search terms report?

Check it weekly while a campaign is new or spending heavily, and monthly once its performance is steady. Choose a rhythm you will actually maintain, and record the date of each review.

What are good negative keywords to start with?

Start with words that describe what you don't sell, which for many shops includes terms like "free", "jobs" or "DIY". Check your own report first, though, because a word that signals junk for one business can signal a genuine buyer for another.

Is it safe to paste my Google Ads data into an AI?

Search terms and costs are business data rather than customer details, but you should still check your AI's data settings and your company's policies first. Remove anything that identifies a customer before you paste it.

Should I let Google's own AI write my ads instead?

You can use both approaches together. Google Ads has its own automated features, which change often, so check Google Ads Help to see what your account currently offers. Your own AI is good at writing from your facts and sorting reports under rules you set. Whichever tool writes the ad, confirm that every claim is true.