Which setup cleans up my search terms without adding negatives I already block?
This is a small UK ecommerce account, a 3D-printing reseller running Search, Shopping and Performance Max. Across three search-term-cleanup prompts, AdTAO wins two and ties one. The win that matters: on the negative-keyword job, running the prompt literally would add 41 negatives, but 19 of them your account already blocks, so AdTAO has you add 22 and skip the duplicates. It also wins on effort (fewest steps to find under-funded converting terms) and ties plain classification, where a basic setup matched it. AdTAO beats Google's official MCP on all three. Every prompt below shows the question, a four-setup comparison, a plain judgement, and the real output.
Measured run, taken from the real transcripts and redacted, with wording simplified to plain English for reading (not sample cells). Redactions use [account], [CID], [amount]. Screenshot/CSV setup was not run and is labelled as such.
Pack verdict: AdTAO wins 2 of 3. On the negatives job it stopped you adding 19 duplicate negatives your account already blocks (you add 22, not 41); it wins on effort finding under-funded converting terms; and it ties plain classification. AdTAO beats Google's official MCP on all three. Screenshot/CSV not measured.
Cem Atik / Harucon Ventures →, from the public "60 Claude Prompts for Ecommerce Google Ads" library. The search-term-cleanup intents below are drawn from that pack.
Which setup classifies search terms into buyer / researcher / freebie / wrong-product / branded?
This job ends in a tie. AdTAO and both API-based setups all sorted the search terms into buckets with spend per bucket, and no setup clearly led. AdTAO named the most real terms from the account, but a plain setup matched it closely on this job, so we score it a tie, not a win. A screenshot cannot sort terms it cannot read.
Full prompt (as tested)
"Look at the search terms for my account [CID]. Classify the significant terms into Buyer (ready to purchase), Researcher (comparing, not buying yet), Freebie (looking for free, cheap, DIY, or coupon), Wrong product (searching for something I do not sell), and Branded. Give me a table with a running spend total per bucket, and tell me what percent of my budget went to non buyers."
Results comparison (this prompt only)
Setup
What it did
Judgement
AdTAO MCP
Full bucket table with spend share and a branded-vs-buyer split, every term taken from your real search terms report.
Partial
A tie with the plain setup on this job. It named the most real terms from the account but did not clearly lead, so this is a tie, not a win.
Google official Ads MCP
Produced a competent bucketed table from raw GAQL. The classification job does not need peer context.
Partial
Produced a competent bucketed table from raw GAQL. The classification job does not need peer context.
Bare LLM
Same bucketing from raw API access; slightly more scratch-work to assemble.
Partial
Same bucketing from raw API access; slightly more scratch-work to assemble.
Screenshot / CSV only
Not measured in this run
Not measured
No read path into live account data
Anonymised actual results (nothing hidden)
Redactions: [account], [CID], [amount]. Every measured setup's output is shown below, redacted and simplified to plain English for reading, with no behaviour hidden behind the summary. Screenshot/CSV was not run.
…Smaller long-tail terms not shown individually are included in bucket totals where classifiable.*
---
### 💸 What % of Budget Went to Non-Buyers?
| Non-Buyer Buckets | Spend (£) | % of Total |
|---|---|---|
| 🔬 Researcher | [amount] | 9.3% |
| 🆓 Freebie | [amount] | 9.0% |
| ❌ Wrong product | [amount] | 1.4% |
| **Total Non-Buyer** | **[amount]** | **🔴 19.7% of budget** |
> **Nearly 1 in 5 pounds spent is going to people who are not buyers.**
---
## 🎯 Priority
Bare LLM Partial (5 tool calls)
…|
|---|---|---|---|
| h2c | [amount] | 1 | 0.0
Screenshot / CSV only [not measured]
Not run in this pack. A screenshot/CSV paste cannot query the account.
Prompt 2 of 3
Which setup builds a ready-to-paste negative-keyword list with recovered spend?
Run this prompt literally and you would add 41 negatives. AdTAO checked the list against the negatives your account already blocks and found 19 of the 41 were already handled, so it skipped them: you add 22, not 41, with no duplicates cluttering your lists. It also checked each of the 22 against results from accounts like yours, to be sure none of them earn money elsewhere before you block them (none did), and against the keywords you are actively bidding on (no clashes). The API-only setups built a reasonable list from this one account, but they cannot see your existing negatives or results from similar accounts, so they cannot tell you a term is safe to block.
Full prompt (as tested)
"From the search terms in my account [CID], build me a ready to paste negative keyword list. Group it into: wrong product, free and cheap seekers, DIY and how to, competitor names to exclude, and irrelevant modifiers. For each group tell me the total spend those terms consumed over the period so I can see what I am recovering."
Results comparison (this prompt only)
Setup
What it did
Judgement
AdTAO MCP
Checked the list against the negatives you already block (19 of 41 skipped), against results from accounts like yours (none earn elsewhere), and against your active keywords (no clashes). You add 22, not 41.
Win
Only setup that checks your existing negatives and results from similar accounts before recommending, so you do not re-add negatives you already have.
Google official Ads MCP
Reasonable grouped negative list from this account's own zero-conversion terms; no check against your existing negatives or similar accounts.
Partial
Reasonable grouped negative list from this account's own zero-conversion terms; no check against your existing negatives or similar accounts.
Bare LLM
Similar list from this one account; no view of your existing negatives or of similar accounts.
Partial
Similar list from this one account; no view of your existing negatives or of similar accounts.
Screenshot / CSV only
Not measured in this run
Not measured
No read path into live account data
Anonymised actual results (nothing hidden)
Redactions: [account], [CID], [amount]. Every measured setup's output is shown below, redacted and simplified to plain English for reading, with no behaviour hidden behind the summary. Screenshot/CSV was not run.
AdTAO MCP Win (3 tool calls)
# Negative Keyword List, Account [CID]
Ready to paste | Last 30 days | 22 new negatives identified
> Audit summary before we start:
> - 22 negatives ready to add (detailed below)
> - 19 proposals already covered by your existing negative lists (skipped, no duplicates)
> - 0 terms we would warn you not to block because they earn for accounts like yours (safe to block all 22)
> - 0 existing negatives conflict with live enabled keywords (no friendly-fire risk)
## Group 1, Wrong Product
Terms that landed on your ads but describe a different product entirely.
| Negative Keyword | Match Type | Wasted Spend |
|---|---|---|
| [product] | Exact | [amount] |
| [branded query] | Exact | [amount] | …
Google official Ads MCP Partial (5 tool calls)
…mount] | 232 | 0 | [brand] campaign - generic educational query |
| sls 3d printer | Broad Match | [amount] | 44 | 0 | [brand] Ad - duplicate, still zero conversions |
| sls printer | Broad Match | [amount] | 76 | 0 | [brand] - generic, no conversion |
| [account] [brand] | Phrase Match | [amount] | 170 | 0 | [brand] campaign - brand+product but zero conversions, likely spec-checking |
| [brand] scanner | Phrase Match | [amount] | 208 | 0 | [brand] campaign - zero conversions despite brand specificity |
**Recommended broad DIY/How-To modifier negatives to add account-wide:**
| Negative Keyword | Match Type |
|---|---|
|
Bare LLM Partial (4 tool calls)
…earing in "UK - Search - [account]" with 123 impressions, 0 clicks - duplicate irrelevant modifier |
**Group 5 Total: [amount] spend (but impression-share drain - add proactively)**
---
## ✅ Ready-to-Paste Master Negative List
Copy the block below and paste directly into **Google Ads → Shared Library → Negative Keyword Lists** (or apply at campaign level):
```
== Group 1: Wrong product ==
[product]
[product]
[branded query]
additive x
== Group 2: FREE & CHEAP SEEKERS (Proactive) ==
free
free download
free software
cheap
cheapest
low cost
budget
open source
diy filament
free stl
free 3d model
== Group
Screenshot / CSV only [not measured]
Not run in this pack. A screenshot/CSV paste cannot query the account.
Prompt 3 of 3
Which setup finds converting terms that are starved of budget inside broad campaigns?
All three setups found the under-funded converting terms, so the finding itself is a tie. AdTAO wins on effort: it reached the same answer in the fewest steps, while the API-only setups ran many more queries to assemble the same picture. The terms convert well but sit inside a broad ad group, losing impression share to lower-value keywords that share the same budget; you fix it by splitting them into their own ad group or lifting the bid so each earner can spend.
Full prompt (as tested)
"In my account [CID], find the products and search terms that convert well but are starved of budget because they are buried inside a broad campaign or ad group. These are what I should promote and fund. List them, show the evidence, and tell me the structure to give each one room to spend."
Results comparison (this prompt only)
Setup
What it did
Judgement
AdTAO MCP
Found the under-funded converting terms and proposed a structure to fund them, in the fewest steps.
Win
Win on effort: it reached the same finding in the fewest steps. The finding itself is a tie across the three setups.
Google official Ads MCP
Reached a comparable finding but took markedly more queries to get there.
Partial
Reached a comparable finding but took markedly more queries to get there.
Bare LLM
Same finding, the most steps of the three.
Partial
Same finding, the most steps of the three.
Screenshot / CSV only
Not measured in this run
Not measured
No read path into live account data
Anonymised actual results (nothing hidden)
Redactions: [account], [CID], [amount]. Every measured setup's output is shown below, redacted and simplified to plain English for reading, with no behaviour hidden behind the summary. Screenshot/CSV was not run.
AdTAO MCP Win (11 tool calls)
…The single ad group "Branded keywords" contains all traffic. Inside it, high-value terms are competing with low-value ones for the same budget:
| Search Term | Clicks | Conv. Value | Convs | Cost | ROAS | Status |
|---|---|---|---|---|---|---|
| **[branded query]** | 32 | **[amount]** | 4.5 | [amount] | **18.6x** | 🔴 Buried |
| **[account]** | 65 | [amount] |
Google official Ads MCP Partial (16 tool calls)
…Inside this one campaign, two products are dramatically outperforming everything else but are competing for budget against 15+ other products.
### The Star Products Being Starved
| Product | Item ID | Conversions | Conv. Value | Spend | Cost/Conv | Impressions |
|---|---|---|---|---|---|---|
| **[account] [brand]** | `[SKU]` | **12.64
Bare LLM Partial (17 tool calls)
…It is also the only Search campaign covering [account], meaning there is no other campaign picking up the slack.
### The Specific Search Terms Being Starved
| Search Term | Status | Impressions | Clicks | Convs | Conv. Value | Cost |
|---|---|---|---|---|---|---|
| **[account] uk** | Added ✓ | 33 | 4 | **2.24** | [amount] | [amount] |
| **
Screenshot / CSV only [not measured]
Not run in this pack. A screenshot/CSV paste cannot query the account.
Where did AdTAO win on this pack, and what is that worth?
→
P2, the negatives you would have re-added. Value (time saved and consistency): running the prompt literally adds 41 negatives; AdTAO found 19 already blocked in your account and skipped them, so you add 22, not 41, with no duplicates. It also checked each of the 22 against results from accounts like yours (none earn elsewhere) and against your active keywords (no clashes). A single-account setup cannot make those checks, so it can have you re-add negatives you already own, or block a term that earns for advertisers like you. You paste the final 22 into your Shared Library negative keyword lists, grouped by match type (exact, phrase and broad match), and check each against your search terms report first.
→
P3, the same answer in less work. Value (time saved): all three setups found the under-funded converting terms, so the finding is a tie, but AdTAO reached it in the fewest steps and showed the real clicks and revenue for each term from your account along the way. These are converting search terms sitting under-funded inside a broad ad group; you act by splitting the ad group or lifting the budget so each earner can spend.
Where did AdTAO lose or only partially meet the pack?
On the classification job, a basic setup with raw API access matched AdTAO: it named about as many real terms from the account. Classification is a straight read of your search terms report, so every good setup gets it right, and a tie there is the expected result, not a loss. Where we compare against accounts like yours, we keep the exact cross-account numbers off the page to protect those advertisers, and we would do the same for yours.
What method and caveats apply?
Model: claude-sonnet-4-6 on every setup. Setup: scripted API (not consumer apps). Run dates: 2026-07-27 to 2026-07-28. Single account: a small UK 3D-printing reseller running Search, Shopping and Performance Max campaigns, low spend band. Setups measured: AdTAO MCP, Google official Ads MCP, bare LLM with raw GAQL. Screenshot/CSV not run. Anonymisation: account names, customer IDs, and exact spend redacted to [account] / [CID] / [amount].
TLDR
Three search-term prompts on a small UK 3D-printing store. The one that matters: run the negatives prompt literally and you add 41 negatives, but 19 are already blocked in your account, so AdTAO has you add 22 and skip the duplicates, and it confirmed none of the 22 earn money for accounts like yours. AdTAO also wins on effort finding under-funded converting terms, and ties plain classification. It beats Google's official MCP on all three. Every setup's output is shown, redacted and simplified to plain English.
FAQ
Does this page hide raw results behind a summary?
No. Every prompt intent includes an anonymised actual-results block for each measured setup, in HTML text.
Which setups were measured?
AdTAO MCP, Google's official Ads MCP, and a bare LLM with raw Google Ads API access. Screenshot/CSV was not run and is marked as not measured.
Is this the same as the search-term-analysis category page?
No. The category page rolls up the job. This page measures one creator pack, prompt by prompt, with anonymised outputs.
Will this add negative keywords I already have?
No. AdTAO reads your existing negative keyword lists first. On this account it found 19 of the 41 proposed negatives were already blocked, so it skipped them and you add 22. A basic setup with no view of your lists would have you re-add all 41. You can confirm this in your shared negative keyword lists in the Google Ads UI.
How much wasted spend does it find, and which metrics does it use?
It works from your own numbers in the search terms report: spend, clicks, conversions, conversion value, CPC and ROAS, by search term and by campaign. The negative-keyword list covers terms with clicks but zero conversions over the last 30 days. The dollar figure is small here because this is a small test account; the same list scales with your budget. Every term traces back to a row you can filter to in your search terms report (set conversions to 0).
What does the buried-winners prompt change in my account?
It finds converting search terms and products that are starved of budget because they sit inside a broad campaign or ad group, and it proposes a structure that gives each one room to spend. You act on it in the Google Ads UI by splitting the ad group or raising the budget on the terms with strong conversion rate and ROAS.