The Amazon search term report is the only place Amazon shows you the exact queries customers typed before clicking your ads. Most sellers download it, scroll through a few thousand rows, fix one or two obvious problems, and close the spreadsheet.
The difference between accounts that compound and accounts that bleed is not access to this data — everyone has it. It is whether the team runs a consistent workflow on it: the same classification rules, the same thresholds, the same cadence, every week.
This guide gives you that workflow. It covers how to pull the report, how to classify every search term into one of five actions, how to harvest and negate correctly, and how to turn the whole process into a repeatable system instead of a monthly spreadsheet marathon.
Quick Answer
An Amazon search term report workflow is a repeatable process for reviewing the customer queries that triggered your ads, classifying each term as harvest, negate, monitor, protect, or ignore, and applying those decisions on a fixed cadence. Converting terms that meet your efficiency threshold get promoted into exact match campaigns; terms with significant spend and no sales get added as negative keywords; low-data terms are documented and revisited. Run the workflow weekly — Amazon retains search term data for only about 60 days, so skipped reviews create permanent blind spots.
Who This Is For
This workflow is written for three readers:
- Sellers and brand operators running their own Sponsored Products campaigns who want a process instead of ad-hoc cleanup.
- Agency account managers who need the same review standard applied across every client account.
- PPC leads who already review search terms but want to formalize thresholds, approvals, and documentation so decisions stop depending on who did the review.
What the Search Term Report Is (and Is Not)
The search term report lists every customer query that triggered at least one click on your ad, with impressions, clicks, spend, sales, and orders for each.
One definition matters before anything else: keywords are not search terms. A keyword is what you bid on. A search term is what the customer actually typed. If you bid on the broad match keyword “stainless steel water bottle” and a shopper searches “insulated water bottle 32 oz pink,” the report shows the latter — and that is the string you paid for.
Three practical constraints shape the workflow:
- Retention is roughly 60 days. Data older than that is gone permanently, which is why download cadence is part of the workflow, not an afterthought (Amazon Ads reporting documentation).
- Low-volume queries are bundled into an “other” row, which can hide a meaningful share of spend. You optimize what is visible, which makes managing the visible terms well even more important.
- Coverage varies by ad type. Sponsored Products gives the most complete search term data; Sponsored Brands has its own report. Verify current console options before building automation around them.
The Classification Framework: Five Actions for Every Term
Every search term in the report should end the review in exactly one bucket. This is the core of the workflow — it removes the “stare at 2,000 rows” problem because each row has a rule-based destination.
| Classification | Meaning | Typical threshold (adjust to your margins) | Action |
|---|---|---|---|
| Harvest | Query converts at or better than target efficiency | 2–3+ orders and ACOS at or below target | Add as exact match keyword in a manual campaign; negate in source |
| Negate | Query spends without acceptable sales | 15–20+ clicks with 0 orders, or ACOS far above target | Add as negative exact (irrelevant themes: negative phrase) |
| Monitor | Early signal, not enough data | Under ~15 clicks, some engagement | Document, revisit next cycle |
| Protect | Strategic term (brand, hero product, defensive) | Judgment call, not threshold | Keep active with bid guardrails regardless of short-term ACOS |
| Ignore | Irrelevant noise, negligible spend | A few impressions or clicks, immaterial cost | No action |
ACOS — Advertising Cost of Sales — is ad spend divided by ad-attributed sales. Most sellers target somewhere between 15% and 30%, but the right target depends on your margin and whether the campaign’s job is profit or ranking. The thresholds above are practitioner conventions, not Amazon rules; set them once, write them down, and apply them consistently.
Two threshold mistakes cause most of the damage:
- Negating too early. A term with 8 clicks and no orders is not a verdict — it’s a small sample. Premature negation kills keywords that would have converted by click 25.
- Harvesting without negating the source. When you promote a term to exact match, add it as a negative exact in the campaign that discovered it. Otherwise both campaigns serve the same query and you compete against yourself.
Step-by-Step Weekly Workflow
This is the full process. Run it weekly; monthly is too slow, because a wasted-spend term can burn its budget for four weeks before you catch it.
- Pull the data. Seller Central → Advertising → Measurement & Reporting → Sponsored Ads Reports. Create a Sponsored Products search term report, summary view, last 30 days (60 for low-volume accounts). Download and archive it — remember the 60-day retention window.
- Aggregate duplicates. The same search term often appears under multiple campaigns and match types. Pivot by customer search term so each query has one consolidated row of clicks, spend, orders, and sales.
- Apply your thresholds. Filter into the five buckets using the classification table. This step should be mechanical — if you’re debating individual rows, your thresholds aren’t specific enough.
- Process harvests. Add each harvest term as exact match in the appropriate manual campaign, set the starting bid near its observed CPC, and negate the term in its source campaign.
- Process negations. Add negative exact for bleeders (real queries that don’t convert for you). Use negative phrase only for whole themes that will never be relevant — it blocks every query containing the phrase, so it’s a wider blast radius.
- Log monitor and protect decisions. Write down which terms you’re watching and why certain expensive terms are deliberately protected. Next week’s reviewer (including future you) needs this context.
- Summarize the run. Terms harvested, terms negated, spend addressed, and anything escalated. This record is what makes week 12 smarter than week 1.
Example: One Workflow Run
Input. A 30-day Sponsored Products search term report for a kitchenware brand. Target ACOS: 25%. After aggregation: 1,840 unique search terms.
Workflow. Apply the five-bucket rules: harvest = 2+ orders at ≤25% ACOS; negate = 20+ clicks with 0 orders or ACOS above 75%; monitor = under 15 clicks; protect = 4 brand terms flagged by the strategist.
Expected output. Something like: 14 harvest candidates (e.g., “silicone spatula set heat resistant” — 6 orders, 18% ACOS, currently matched via broad), 23 negation candidates (e.g., “spatula city” — 31 clicks, $42 spend, 0 orders), ~200 monitor terms documented, 4 protected brand terms untouched, the rest ignored.
Business decision. The 14 winners move into exact match campaigns with source negation; the 23 bleeders stop spending immediately; roughly $300–400/month of waste is redirected toward terms with proven conversion. The decisions and reasoning are logged so next week’s run starts from a known state.
Tools and Data Needed
- Sponsored Products search term report (weekly download)
- Campaign and ad group structure map — where harvests should land
- Target ACOS or TACOS by product or campaign objective
- Negative keyword lists per campaign (to avoid duplicate negation)
- A decision log — spreadsheet tab or, better, a workflow tool that records actions automatically
Common Mistakes
- Reviewing monthly instead of weekly. Slow cadence plus 60-day retention means you act on stale data and lose history.
- Negating on thin data. Under ~15 clicks is a coin flip, not a signal.
- Forgetting source negation after harvest. Causes self-competition and muddies your data.
- Using negative phrase too aggressively. One careless phrase negative can block dozens of converting long-tail queries.
- Treating thresholds as folklore. If your rules live in one person’s head, every reviewer produces different decisions. Write the SOP down.
- No decision history. Without a log, you re-litigate the same terms every week and can’t explain account changes to a client or teammate.
Where the Manual Workflow Breaks Down
Run manually, this process works — for one account, with one disciplined operator, for a while. It breaks down in predictable places:
- Volume. Hundreds of campaigns across several accounts means tens of thousands of rows weekly. The pivot-table ritual quietly becomes a full day of work, and steps get skipped.
- Consistency. Two account managers with the same SOP will still classify edge cases differently. Standards drift, and nobody notices until performance does.
- Auditability. A spreadsheet shows what changed, not why. When a client asks why a keyword was negated in March, the answer is usually a shrug.
- Cadence under pressure. The week everything is busy is the week the review gets skipped — and with a 60-day window, skipped weeks are unrecoverable.
You can use a general-purpose LLM to help here — Claude can help design the workflow logic, classify a pasted report, or draft your SOP. But a chat session doesn’t remember last week’s decisions, enforce thresholds, or route actions through approval. That’s a workflow problem, not an intelligence problem — a distinction we’ve argued in depth in Why LLMs Should Build Amazon Ads Workflows, Not Run Them.
Where Trellis Fits
Trellis Amplify is built for exactly this gap: it turns the SOP you just read into a saved, executable workflow. You define the classification rules and thresholds in plain English, the workflow ingests search term data on schedule, classifies every term into harvest / negate / monitor / protect / ignore, and produces an action list with reasoning. A human approves before anything changes, and every decision is recorded — so the review runs the same way across every account, every week, regardless of who is at the keyboard.
If you currently run this review in spreadsheets, the migration path is short: your thresholds and rules are the workflow.
See how Amplify turns your search term SOP into a repeatable workflow →
For the broader context on AI-assisted PPC operations, see AI for Amazon Ads and our guide to the Amazon Ads MCP Server.
FAQ
How far back does the Amazon search term report go? About 60 days. Older data is permanently unavailable, so download and archive reports at least every 30 days.
How often should I review my search term report? Weekly for active accounts. Weekly cadence catches wasted-spend terms before they burn a month of budget and keeps you inside the retention window.
How many clicks before I negate a search term? A common convention is 15–20+ clicks with zero orders, adjusted for your price point and conversion rate. Higher-priced products warrant more patience; negating under 15 clicks is usually premature.
Should I use negative exact or negative phrase? Negative exact for specific queries that don’t convert for you. Negative phrase for entire themes that will never be relevant — used sparingly, because it blocks every query containing the phrase.
What is keyword harvesting on Amazon? Moving proven, converting search terms out of discovery campaigns (auto, broad) into controlled exact match campaigns — and negating them in the source campaign so the two don’t compete.
What does the “other” row in the report mean? Amazon groups low-volume queries it doesn’t report individually. It can represent a significant share of spend; if it’s unusually large, restructure discovery campaigns to surface more individual terms.
Does the search term report cover organic traffic? No — it only includes queries that triggered paid ad clicks. For organic and market-share data, brand-registered sellers can use Brand Analytics Search Query Performance.
Can AI run this workflow for me? AI is best used to build and operate the workflow under human review: classify terms against your thresholds, draft the action list, and document decisions. Letting an unconstrained AI apply bid and keyword changes without approval steps removes the governance that makes the process safe.
Conclusion
A search term report only creates value when a consistent process acts on it. Define your five-bucket thresholds once, run the review weekly, always negate the source after harvesting, and log every decision. Then stop doing it by hand: encode the same rules into a workflow so the review runs on schedule, survives busy weeks, and produces the same standard across every account.
Next step: take the SOP from this article and turn it into an executable workflow with Trellis Amplify.
