Move search terms from Auto to Manual

On this page you can find the search terms that perform well in your Auto campaigns and build candidates to carve them out as keywords in your Manual campaigns. Use it to concentrate winning search terms into Manual, playing to the division of labor where Auto is for discovery and Manual is for precision.

About this page

Item	Details
What you can do	List candidates to promote from an Auto campaign’s search terms into Manual-campaign keywords, complete with a recommended match type and recommended bid
Applies to	Sponsored Products (SP)
Data & connection needed	The connected account’s search-term and campaign performance data (no extra setup required)
Scope	Analysis and proposal only (read-only). The prompts on this page only build promotion candidates and a migration plan; they do not change ad settings. The actual additions to Manual and the exclusions on the Auto side are carried out through the approval flows on the related pages

Terms used on this page

• Auto campaign (automatic targeting) … a serving method where Amazon automatically chooses the search terms. It plays the “discovery” role of picking up search terms you have not noticed yet.
• Manual campaign (manual targeting) … a serving method where the advertiser specifies the keywords. It plays the “precision” role of concentrating bids on terms that clearly get bought.
• search terms … the words shoppers actually type into Amazon.
• keywords … the words the advertiser registers on the ad. They are distinct from search terms. Registering a search term that performed well in Auto as a keyword in Manual is what we call “promotion”.
• match type … the matching condition of a keyword. Three kinds: Exact / Phrase / Broad.
• negative keywords … a setting that registers search terms you do not want your ad to show against, removing them from the serving pool.
• overlapping delivery … the state where the same search term is picked up by both Auto and Manual, making your own campaigns compete on bids internally. You prevent it by adding a promoted search term as a negative keyword on the Auto side.
• CVR (conversion rate) … the share of clicks that led to a purchase. Higher is more efficient.
• ACoS … Advertising Cost of Sales: ad spend ÷ ad-attributed sales. Lower is more efficient.

How Auto and Manual divide the work

Auto plays the discovery role of “picking up search terms you have not noticed”, and Manual plays the precision role of “concentrating bids on search terms that clearly get bought”. The standard play is to move winning search terms out of Auto into Manual, and to add the same search term as a negative keyword on the Auto side to prevent overlapping delivery.

Watch out for the purchase-count and click-count fields

Search-term and keyword performance data contains several metrics with similar names.

• For the purchase count, use the order count. It is a different value from “units sold” (which goes up when several units are bought in the same order), so don’t conflate them.
• Click-count and purchase-count thresholds do not work as server-side filters on many tools. Rows below the value you specify can still come back, so re-filter against the actual data after fetching.
• There is a limit on how many rows of search-term data can be retrieved. Always reconcile the number of rows you retrieved against the overall total, and if you have hit the limit, widen the fetch range before drawing conclusions.

When to use this

• You want to carve search terms that are selling in an Auto campaign out into keywords in a Manual campaign
• You want to move high-converting search terms into Manual and, at the same time, exclude them on the Auto side to prevent overlapping delivery
• You want to sort, once a month, the high-converting terms not yet promoted from the terms already in Manual whose bids you want to strengthen

Before you run

• Connected account … if you handle multiple accounts, check first that the target account is correct.
• Target period … “the most recent completed week”, “the past 30 days”, and so on. Same-day and previous-day data can still be unconfirmed, so a recent completed period is safest.
• Destination and exclusion campaign names … in Prompt 2, replace the destination Manual campaign name and the Auto-side campaign name that receives the exclusions with your actual campaign names.
• If you register a target ACoS and the like in Set targets for sales, ad spend, and ACoS, the judgment criteria are used automatically.

Prompt cheat sheet

Goal	What you learn / what gets built	Changes	Prompt
Find Manual-promotion candidate search terms with standard conditions	A list of promotion candidates with recommended match type and bid	No change	Prompt 1
Build a plan to move high-converting search terms into Manual and exclude them on the Auto side	An end-to-end move + exclusion plan	No change	Prompt 2
Sort monthly high-converting terms into promotion and bid strengthening	A list classifying promotion and bid-increase candidates	No change	Prompt 3

This page goes only as far as “building candidates and plans”

As the “Changes” column of the cheat sheet shows — all “No change” — this page goes only as far as building promotion candidates and a migration plan. Actually adding keywords to Manual is done in Create Sponsored Products (SP) campaigns, and excluding them on the Auto side is done in the approval flow of Add negative keywords (essential to prevent overlapping delivery).

---

Prompt 1: Find Manual-promotion candidate search terms with standard conditions

Lists, with the evidence behind each, the Auto-campaign search terms worth promoting to Manual-campaign keywords.

When to use — when you want to sweep up the search terms performing well in Auto under standard conditions.

What you need — the target period (defaults to the most recent completed week).

Using the most recent completed week of data, detect candidates from the
Auto campaign's search terms to promote into the Manual campaign.
- 30 or more clicks
- 2 or more orders
- CVR of 5% or higher
Include the recommended match type and recommended bid as well.

What you can change — the target period / click count (e.g. 30 or more) / order count (e.g. 2 or more) / conversion rate (CVR, e.g. 5% or higher). Adjust by rewording the numbers as “N or more clicks”, “N or more orders”, “CVR of N% or higher”.

What you get back — a table of search term × recommended match type (Exact / Phrase / Broad) × recommended bid × observed CVR × order count × whether an Auto-side negative-keyword registration is needed.

Changes — analysis and proposal only. Does not change ad settings.

Next steps — additions to Manual are carried out in Create Sponsored Products (SP) campaigns, and Auto-side exclusions in the approval flow of Add negative keywords (essential to prevent overlapping delivery).

Known issue

When detecting promotion candidates, the same search term can appear multiple times (up to three) as a duplicate (being fixed). Duplicates are removed automatically when the change is applied, so there is no operational impact, but the apparent count is higher than the real one.

---

Prompt 2: Build a plan to move high-converting search terms into Manual and exclude them on the Auto side

Identifies the high-converting search terms among your Auto search terms, moves them into the Manual campaign as Exact Match, and at the same time adds the same terms as negative keywords on the Auto campaign — an end-to-end migration plan.

When to use — when you want to concentrate search terms that are clearly getting bought into Manual and clean up the overlapping delivery with Auto in one pass.

What you need — the target period (defaults to the past 30 days), the destination Manual campaign name, and the Auto-side campaign name that receives the exclusions.

From the past 30 days of search terms in Auto targeting campaigns, identify high-converting search terms with 3 or more orders and an ACoS of 20% or lower, and build a Manual promotion plan.

[Procedure]
1. First present the campaign names that actually exist on the Auto side and as candidate Manual promotion destinations. If {{Auto-side campaign name}} and {{destination Manual campaign name}} are specified, use those; if not, show alternatives from campaigns that actually exist. Verify a campaign's existence with a partial-match name search — don't fetch the entire list.
   State the basis for telling Auto from Manual (whether the targeting_type value could be retrieved, or whether it is an inference from the naming convention). If the promotion destination does not contain "Manual" under the naming convention, say so and show the promotion-destination candidates that do exist.
   * Campaign names mix the half-width | and full-width ｜ separators. A name search picks up both via NFKC normalization, but if you split the name and judge it yourself, handle both.

2. Fetch the search term report.
   🔴 You can tell the purchase count from the field name. Don't infer it.
   Each row contains purchases_p2 (order count) and units_sold_clicks_7d_p2 (units sold) as two separate values
   (measured: one row had purchases_p2 442 / units_sold_clicks_7d_p2 450; another had 377 / 388).
   Use purchases_p2 (order count) as the "purchase count" in this prompt, and state in the column name which one you used.
   (The inference that "a row where the order count exceeds the click count is proof that units sold has been mixed in" is no longer needed.)
   🔴 Use _p2 for the period. Every metric comes back in 3 series: _p1 (the immediately preceding comparison period of the same length) / _p2 (the specified period) / _delta.
   The actual two periods are stated in the response's comparison_period, so treat that as authoritative and transcribe it into the output (no separate control check is needed).
   🔴 Secure the fetch volume. The max_pages limit is 100 (100 rows per page = up to 10,000 rows), and it is 20 when omitted.
   Always raise it to the limit. Measured (14 days): upstream total 11,692 / 117 total pages.
   max_pages=10 gives 1,000 rows (8.6%); max_pages=100 gives 10,000 rows (about 85%).
   * min_clicks is not a server-side filter on the search term report (it is applied after fetching).
     In measurements, passing min_clicks=20 left total_results at 11,692. The only thing that increases the fetch volume is max_pages.

3. List the terms meeting the conditions together with their purchase count, ACoS, sales, ad spend, CPC, and source campaign name, sorted by ad spend impact, largest first.
   🔴 Narrow to Auto campaigns after fetching. campaign_id is included on every row (confirmed by measurement), so
   filter the rows by the Auto-side campaign_id you identified in step 1. There simply is no filter argument — campaign_id itself is available.
   State how many rows you kept and how many you excluded.
   Transcribe the response's count fields as they are: total_results / fetched_row_count / filtered_row_count /
   search_term_fetch_truncated / search_term_fetched_pages / search_term_total_pages_available.
   Calculate coverage = fetched_row_count ÷ total_results, showing the numerator and denominator. If it is "not the full set," write that.
   * The upstream fetch order is ad spend descending. This prompt sorts by ad spend impact, so the fetch order matches the objective and any bias from missed rows is small.

4. Build the promotion plan as a table. If a search term is in ASIN form (10 alphanumeric characters starting with B0), promote it via product targeting rather than as an Exact keyword; promote normal terms as Exact Match.
   Also present the proposed Negative Phrase additions on the Auto side, and state both bases for the recommended bid: "the current CPC (cpc_p2)" and "the allowable CPC derived from the target ACoS (AOV × target ACoS = allowable CPA; allowable CPA × CVR = allowable CPC)."
   * acos_p2 / roas_p2 / ctr_p2 / cvr_p2 are all expressed as percentages (measured: on a row with clicks 2,509 / impressions 172,296, ctr_p2 = 1.46). Align the units when working backwards.

5. Check whether the same term already exists at the promotion destination, and state how many you were able to check out of the total.
   🔴 The search_key used for the check is a partial match. Don't take total_results > 0 as evidence that it "exists."
   Fetch the rows and decide based on whether any row's targeting exactly matches the candidate term.
   (search_key is an upstream filter, so there is no retrieval-rate problem at this checking stage. In measurements, total_results dropped from 22,668 to 254.)
   If you cannot check all of them, say so.

6. State in one sentence the effect expected after the promotion is executed (ACoS improvement, cutting wasted bids). Do not forecast numbers.

[Constraints]
- Present the plan only. Execution happens in a separate prompt after approval. Do not fill in data you couldn't retrieve by guesswork.
- Do not include unfinalized days such as today in the period. Check the data ingestion status and cover through the last finalized day. Look at the per-report status, not rollup_status.
- Note that terms with an ACoS under 1% may be a measurement discrepancy. If there are 0 matches, say so and stop.
- The upstream aggregation is cached for about 4 hours, so state that recent changes may not be reflected.

Before executing, always: present the promotion/exclusion list (target term / current home / how it will be treated / expected impact) and do not apply anything to the platform until approved.
After executing: after approval, apply the changes only once the user reviews and approves the proposal, then 14 days later compare ACoS and purchase count with "measuring the effect of a change" and judge Keep/Adjust/Revert. Use the same field (purchases_p2) and the same period length when re-fetching.
* The execution paths for the promotion (adding keywords, adding negatives) go through a different approval flow. The bid-change approval flow is not intended for adding or excluding targets, so check that the execution steps match this operation (adding and excluding targets) before connecting them.

What you can change — the target period / order count (e.g. 3 or more) / ACoS (e.g. 20% or lower) / the destination Manual campaign name (Manual-Brand in the example) / the Auto-side campaign name that receives the exclusions (Auto-Discovery in the example). Replace the campaign names with your actual names, and reword the conditions as “N or more orders”, “ACoS of N% or lower”.

What you get back — a table of the keywords to move × recommended bid × Auto-side exclusion proposals (a proposal to exclude them as Phrase match).

Changes — analysis and proposal only. Does not change ad settings.

Next steps — after approval, additions to Manual are carried out in Create Sponsored Products (SP) campaigns, and Auto-side exclusions in the approval flow of Add negative keywords, in order.

---

Prompt 3: Sort monthly high-converting terms into promotion and bid strengthening

At monthly granularity, identifies the high-converting terms with many orders and sorts them: those not yet in Manual as promotion candidates, and those already in Manual as bid-increase candidates.

When to use — when you want to sort, once a month, the search terms you should promote from the existing keywords whose bids you should strengthen.

What you need — the target month (defaults to the previous month).

For {{target month (e.g., 2026-04)}}, identify high-converting keywords with {{threshold (e.g., 4)}} or more orders.
[Procedure]
1. Run analyze_promotion with the period of {{target month (e.g., 2026-04)}} and min_orders={{threshold (e.g., 4)}} to get the promotion candidates. State the number of rows returned (item_count). This tool does not return the total size of the population, so write only "the number of rows retrieved" — don't write an estimated total (don't fill it in by guesswork).
2. Don't use the returned rows as they are; always reconcile them against the actual data in step 3 before judging. There are 4 reasons.
- The "count" is units sold, not the purchase (order) count, and CVR is calculated as "units sold ÷ clicks." That is why rows with a CVR over 100% appear.
- The filter is an OR condition: a row comes back if it satisfies either "count ≥ min_orders" or "CVR ≥ min_cvr." Rows below min_orders get mixed in even when you specify it.
- min_clicks is not applied to the returned rows. Rows with fewer clicks than the value you specify still come back.
- current_value's "Auto (not registered)" is a label applied uniformly without checking the registration status, and it is attached even to terms already registered as Manual keywords. Don't use it as evidence that something is unregistered.
3. For each candidate term, call list_search_term_performance with the period of {{target month (e.g., 2026-04)}} and search_key=that term, and fetch the actual data.
- Confirm that total_results and fetched_row_count in the response match. If they don't, raise max_pages and re-fetch; if they still don't match, state explicitly that "the full set could not be retrieved."
- Determining registration status: if there is a row whose match_type is Exact/Phrase/Broad and whose targeting matches the term, it is "registered in Manual." If there isn't, it is "not registered (Auto side only)."
- Getting the metrics: use the series for the specified period (suffix _p2). Purchase count = purchases_p2, ACoS = acos_p2, sales = sales_7d_p2, ad spend = cost_p2, clicks = clicks_p2, current bid = keyword_bid. If the same term is split across several rows, sum them before judging.
- purchases_p2 (purchase count = order count) and units_sold_clicks_7d_p2 (units sold) are different things — don't conflate them.
4. Filter and list using the actual data obtained in step 3.
- Keep only terms whose purchases_p2 is {{threshold (e.g., 4)}} or more. State how many you excluded for falling below it.
- Conversion rate cannot be judged for terms with too few clicks, so exclude terms whose clicks_p2 is under 30 and state the number excluded.
- If you exclude terms in ASIN form (10 digits starting with b0/B0), state the number excluded — don't exclude them silently (if you do exclude them, state that "this was excluded on my own judgment").
- Unregistered terms → list them as promotion candidates, together with purchase count, ACoS, sales, and ad spend.
- Terms already registered in Manual → list them separately as bid-increase candidates, together with the current bid (keyword_bid), purchase count, ACoS, sales, and ad spend.
- If there are 0 matches, say so and stop.
[Constraints]
- Do not fill in data you could not retrieve by guesswork. Do not perform write operations.
- Distinguish null, not retrieved, no data, and 0. Do not treat "not retrieved" as 0.
- analyze_promotion generates a proposal awaiting approval (proposal_id) simply by being called. This call is for analysis, so don't run execute_proposal. Discard the proposal with cancel_proposal once it is no longer needed.

Before executing, always: present the list of promotion and bid-increase candidates (target / current value / how it will be treated / expected impact) and do not apply anything to the platform until approved.
After executing: after approval, apply the changes only once the user reviews and approves the proposal, then 14 days later compare ACoS and purchase count with "measuring the effect of a change" and judge Keep/Adjust/Revert.

What you can change — the target month (replace with the month you want to analyze; defaults to the previous month) / the order-count threshold (4 or more is a good baseline for a monthly view; reword as “N or more orders”).

What you get back — a table of keyword × order count × ACoS × existing serving status × recommended action (promote / raise bid).

Changes — analysis and proposal only. Does not change ad settings.

Next steps — promotion candidates are carried out in the approval flow of Create Sponsored Products (SP) campaigns, and bid-increase candidates in that of Adjust bids.

---

Full promotion flow

1. Detect promotion candidates on this page (Prompts 1-3)
   ↓
2. Add keywords to the Manual campaign in
   [Create Sponsored Products (SP) campaigns](/en/prompts/campaign-creation/) (approval flow)
   ↓
3. Register the same search terms as
   [negative keywords](/en/prompts/negative-keywords/) on the Auto campaign (approval flow)
   (prevents overlapping delivery between Auto and Manual)
   ↓
4. Optimize the promoted Manual keywords with
   [Adjust bids](/en/prompts/bid-optimization/)

Related categories

• Create Sponsored Products (SP) campaigns — Adding keywords to a Manual campaign
• Add negative keywords — Auto-side exclusion registration after promotion (prevents overlapping delivery)
• Adjust bids — Optimizing promoted Manual keywords
• Find high-performing keywords and product targeting — Surfacing additional candidates, including promotion candidates
• Analyze high-performing search patterns with N-grams — Anonymization constraints and interpretation guide for Auto-derived search terms