Google Ads search terms and negative keywords: how to shortlist candidates with AI

What is the search terms report, and what is missing from it?

The Google Ads Help Center describes the report as a list of searches that triggered your ad and how those searches performed (checked October 22, 2026). You need to know three limits so you do not draw conclusions from what you cannot see.

1. Not all terms are shown. Google says that, for privacy standards, terms with too little search activity are left out of the report. So the row totals will not necessarily match the campaign spend: the rest is simply not shown, and the AI cannot guess it.
2. Not all conversions. Store visit and store sales conversions are not shown in this report.
3. Campaign type changes the picture. Performance Max campaigns have a separate search terms report (data available from March 2023), and Google advises looking at conversions first there, not clicks. If AI Max for Search campaigns is turned on, you can filter the "Match type" column by the value "AI Max" to separate terms that did not come through your keywords (Google warns that numbers may be lower with this filter). Google says negative keywords still apply with AI Max turned on, but advises waiting at least two weeks after turning on AI Max before making changes such as adding negative keywords. Feature availability and menu names depend on the account and the interface language, so check in your own account. Which other AI features Google Ads has built in is covered separately (Google Ads AI features).

How to export the search terms report and what to keep in the file

1. In your Google Ads account, open Campaigns → Insights and reports → Search terms (in the Lithuanian interface "Kampanijos → Įžvalgos ir ataskaitos → Paieškos terminai").
2. Choose a date range. Take a longer one, for example the last 60–90 days: in one week of data, a single term with 2 clicks tells you nothing.
3. Keep these columns: search term, match type, campaign, ad group, impressions, clicks, cost, conversions. If you have conversion value, keep that too.
4. Click the download button and choose a format (CSV, Excel or Google Sheets).
5. Open the file and review it before copying anything into the AI: delete the account number, campaign names if they contain client or partner names, and rows where people typed their own or someone else's names, phone numbers or addresses into the search. This is rare, but it happens. More on this in the data section below.

If you have thousands of terms, start with the ones you spent the most on: sort by cost and take the first 200–300. The long tail with one click each costs little, and the AI works worse with a list that is too long.

Illustrative example: what you get after exporting

Say a home appliance repair shop in Kaunas (Lithuania's second-largest city) advertises washing machine and dishwasher repair in Kaunas only. Below is a made-up extract from the export (the numbers are illustrative, not real). The searches were in Lithuanian; they are shown here in English.

A few clear cases are visible at a glance, but a real report has hundreds of these rows, all mixed together. This is where the AI saves work: it sorts them into groups.

Prompt: groups, candidates and "unclear"

Three things matter most in the prompt: context about the business (what you sell, where you work, what you do not sell), a requirement to give a reason for every candidate, and a separate "unclear" group. Without that last one, the AI tends to push everything to one side or the other, even though some terms cannot be judged from the wording alone. Replace the parts in square brackets with your own information.

Why ask for cost and conversions next to every candidate: you see straight away whether the AI suggested blocking something that brings customers, and you can check the numbers against your own file. For more on structuring prompts like this, see the guide on how to write prompts (How to write AI prompts).

What the AI returns: an example result

With the example table above, the answer could look roughly like this (shortened):

Unclear: "bosch washing machine error e18": the person has a broken machine and could become a customer if they cannot fix it themselves. It is worth checking whether similar "error code" terms have brought calls over a longer period. Have conversions, do not block: "washing machine repair kaunas", "washing machine repair price", "dishwasher repair kaunas".

This is a good starting point, but not yet a decision. Check every row.

How to check each negative keyword candidate

Go through the AI's list with your export file open next to it, and answer five questions for each candidate.

1. Has this word ever brought a conversion? Filter the file by the word (not just the exact term) and over a longer period. If "used" never converted but "used washing machine repair" did, a broad negative "used" would block the good search too.
2. How much does it actually cost? A candidate that cost €4 over three months is low priority. Deal first with the ones you spend the most on.
3. Is there enough data? A term with 3 clicks and 0 conversions does not yet show it is bad. If the intent is clearly wrong (a job search, another city where you do not work), you can block it even from a small sample. If the intent is unclear, it is better to wait.
4. Will it block a good search? Think about which relevant searches this word could appear in. The Google Help Center warns separately: if a negative keyword matches one of your regular keywords, the ad will not show.
5. Which match type? More on this in the next section. Treat the match type the AI suggests as a suggestion, not a rule.

How negative keyword match types differ from regular keywords

This is the most common source of mistakes, so it is worth knowing what Google says (checked October 22, 2026).

Match types in Search campaigns. A broad negative (e.g. used) blocks a search if it contains all the words of the negative keyword, in any order. Phrase (e.g. "how to replace") blocks it if the words appear in the same order, with other words allowed before and after. Exact (e.g. [used washing machine]) blocks only when the search is exactly these words in the same order, with nothing extra.

No close variants. Regular keywords can also match similar searches, but negatives cannot: Google says negative keywords do not match close variants or other expansions, and you have to add synonyms and singular and plural forms yourself (capitalization and misspellings, according to Google, are handled automatically). In Lithuanian this matters even more because of noun cases: the negative "darbas" (job) will not block the search "darbo skelbimai meistrui" (job ads for a technician), because "darbo" is a different form of the same word. In practice, add the forms you see in your own report.

Adding straight from the report: you select the term and click "Add as negative keyword". Google says that in Search campaigns a keyword added this way is an exact negative by default. That is safe, but it blocks only that one search. If you want to block a whole topic (e.g. all "job" searches), enter a shorter word with broad or phrase match.

Level. When adding from the report you can choose the ad group or the campaign: the keyword is added to the ad group or campaign the term came from. Ad group level suits a term that is wrong for just one group (e.g. "dishwasher" in the washing machine group when dishwashers have their own group). Campaign level suits a term that fits nowhere in that campaign.

Lists. If the same words are wrong for several campaigns (job searches, "free", other cities), it is easier to create a negative keyword list in the shared library and apply it to the campaigns. Google states a limit of 20 lists per account and 5,000 keywords per list. At account level you can add up to 1,000 negative keywords.

Performance Max. Google says negative keywords here apply to Search and Shopping traffic and are added at campaign or account level. To block searches for your own brand, Google recommends brand exclusions rather than negative keywords, because they also cover spelling variants.

Common mistakes

Small sample. A one-week report and terms with 1–2 clicks are noise. The AI will group it just as confidently as a large sample, so thinking about whether there is enough data is your job.

Blocked converting traffic. This usually happens with a broad negative that is too short, such as "price" or "cheap", because these words also appear in searches by people ready to buy. After adding negative keywords, compare conversions a few weeks later with the previous period of the same length.

A keyword suggested by AI is not proven demand. If you ask, the AI will most likely also suggest new keywords to "add". Those are guesses, not data. The prompt above deliberately does not allow this. Check demand for new keywords in the Google Ads Keyword Planner and with the campaign itself.

Negative keywords instead of settings. If terms for other cities show up, check the campaign's location settings first. A negative "vilnius" or "klaipėda" is an extra measure, not a replacement.

A report without dates in mind. If you have just added a negative keyword and the term is still visible, check whether the report's date range includes days before the change. Change history shows the date of the change.

What data should you not paste into an AI tool?

Search terms are usually anonymous, but people sometimes type first names, surnames, phone numbers or addresses into the search box. Delete those rows before copying: the analysis does not need them. Also do not paste your Google Ads account number, login details, customer lists or conversion data with personal information. If campaign names contain partner or client names, replace them with neutral ones (A, B, C).

In a company account, check whether your ChatGPT or Claude plan allows the data you enter to be used for model training, and turn it off in the settings if needed.

When this review repeats every month, you can combine it with a general review of campaign results. How to analyze the whole Google Ads report with AI and separate observations from guesses is covered separately (Google Ads report analysis with AI). Manual copying can also be replaced with a technical integration (Google Ads API, n8n), but that is a separate project, and a person should still make the decision on every negative keyword.

Sources

Google Ads Help: About the search terms report: what the report is, low-activity terms hidden for privacy, downloading, incompatible conversions.

Google Ads Help: Negative keyword ideas from the search terms report: adding from the report, choosing ad group, campaign or list, exact match by default.

Google Ads Help: About negative keywords: match types, close variants, synonyms and plurals, account-level negative keywords.

Google Ads Help: Add negative keywords to campaigns: conflicts with regular keywords, list and account-level limits.

Google Ads Help: Search terms report for Performance Max: data from March 2023, conversions first, negative keywords at campaign or account level.

Google Ads Help: Negative keywords in Performance Max campaigns: applied to Search and Shopping traffic, brand exclusions.

Google Ads Help: AI Max for Search campaigns FAQ: negative keywords apply with AI Max turned on.

Google Ads Help: Reporting in AI Max campaigns: the AI Max filter in the search terms report, waiting at least two weeks before making changes.