The Best Keyword Tool Is Your Search Terms Report

The most underused keyword tool in Google Ads is sitting inside the account: the search terms report. It is the log of real queries that matched your ads, each row stamped with what you paid and what converted. External tools estimate demand you might capture; this report records demand you already bought. For an account with history, mining it beats buying another subscription — and a connected AI removes the one excuse for not doing it, which was the spreadsheet work.

Why paid-for queries beat estimated volumes

Every third-party keyword tool answers a hypothetical: how many people might search this, and what might a click cost. The search terms report answers something better: which exact queries did trigger your ads, what you actually paid, and which ones turned into conversions in your account, with your landing pages, at your prices.

That difference matters at decision time. A tool cannot tell you that "industrial dehumidifier hire" converts for you while "dehumidifier rental" does not — the words are near-synonyms, the buyers are not. Your report can, because both queries already cost you money. Keyword research from your own data is not research at all in the speculative sense. It is accounting.

The catch has always been effort. The report is long, repetitive and messy, and the interesting rows hide among thousands of trivial ones. That is clerical work — exactly the class of work an AI agent connected to your account does without complaint.

The expansion prompt: finding queries you never bid on

The core harvest question is simple: which search terms converted, or showed strong intent, without a matching keyword in the account? Connected over MCP, the AI reads the report directly — no export, no pivot table. A prompt that works:

Pull search terms for the last 90 days across all Search campaigns. List every query with at least one conversion, or at least ten clicks, that is not already a keyword anywhere in the account in any match type. For each, show cost, clicks, conversions, the triggering keyword and its ad group. Rank by conversions, then by cost. Flag any query that appears in more than one campaign.

The last line matters more than it looks: a query matching in several campaigns is usually a sign your structure is competing with itself, and the harvested keyword should land wherever you decide that traffic belongs — with negatives fencing it out of everywhere else.

What comes back is a shortlist with evidence attached, not a hunch. Each candidate arrives with the numbers that justify it, which makes the next two decisions — match type and placement — quick.

Which match type should a harvested keyword get?

A harvested query earns its match type from the evidence behind it. The useful default: the more conversions a specific query has, the more precisely you should own it.

Signal in the report	Action	Match type
Repeat conversions on one exact query	Add it as its own keyword	Exact
Conversions spread across close variants of a phrase	Add the common phrase	Phrase
A converting theme with scattered wording	Add the head term, police weekly	Broad
Clicks and cost, zero conversions, wrong intent	Exclude it	Negative
Clicks and cost, zero conversions, right intent	Leave it matched, watch it	None yet

Two notes on that table. Exact match for proven converters is about control: when a query is identical to an exact keyword, that keyword is prioritised, so you decide the bid, the ad and the landing page for your best traffic. And a broad addition is a commitment to reading what it matches into — harvesting into broad without a query review habit just reopens the leak you were closing.

Putting new keywords into the right ad groups

A harvested keyword inherits its home from the ad you want to show. Before adding, ask the AI to check fit: does an existing ad group's RSA actually speak to this query? If yes, add the keyword there. If no ad group does, that is not a keyword decision any more — it is a small build: new ad group, tight theme, headlines written against the query language you just harvested.

This is where doing the work in one conversation pays. The same session that found "industrial dehumidifier hire" can draft the RSA for it, create the ad group, add the keyword at exact, and add the phrase negative to the general ad group so traffic stops splitting. With a write-capable connector each of those changes is shown to you for approval before it lands — additions are cheap to approve because new keywords and paused drafts cannot spend until the structure around them goes live.

The monthly harvest-and-prune loop

Harvesting once is a cleanup. The compounding value is a loop — and because the prompts are reusable, the loop costs about twenty minutes a month:

1. Harvest — run the expansion prompt over the last 90 days. Add what the evidence supports, at the match type the table above suggests.
2. Prune — ask for last month's additions and their performance since. Keywords that matched nothing, or matched garbage, get cut or negated. Negative keywords are half of every harvest.
3. Reconcile — ask which queries now match a different keyword than last month. Match drift is how carefully placed traffic quietly migrates to the wrong ad group.
4. Log — keep the session summary. Next month's loop starts by checking what last month's changes did, which is how the loop learns.

For the deeper version of step 1 — decomposing queries into recurring fragments instead of whole terms — run an n-gram analysis over the same report; it catches waste patterns that whole-query reading misses.

Where external keyword tools still earn a place

Your own data has one blind spot: it only contains demand your current keywords could reach. Three cases still justify an outside tool — a genuinely new market or product line where you have no history; competitor-gap questions, where you want terms rivals rank or bid on; and zero-based planning for a new account, where there is no report to mine yet.

Even then, sequence matters. Outside tools propose; your account disposes. Every imported idea should graduate through the same evidence bar the harvest loop applies to your own queries — spend, then judge, then scale. The report is where keyword research ends up even when it starts elsewhere.

There is also a structural reason the report keeps gaining on the tools: matching keeps loosening. Broad match, AI-driven expansion and conversational query drift all mean your account now samples demand adjacent to your keywords automatically — the search terms report has quietly become a discovery instrument, recording expansion experiments you did not design. A tool guesses what people might search; your account is already running the survey, with a price on every response. Mining it first is not frugality. It is just reading your own results before commissioning someone else's.

If the mining itself is the bottleneck, that is the part you can hand over today: connect the account and run the expansion prompt as written. Start a free pilot — seven days, full toolset, up to five accounts, no card.