Suggest keywords and targets, or tell the engine hands off
How to feed the engine keywords, ASINs and negatives through the agent, and how to carve out entities the engine must leave alone — permanently, or under a bid ceiling.
TL;DR: Ask the agent to add keywords, product targets or negatives and it shows you exactly what it will send to Amazon before anything goes through. Anything the agent sets is claimed for seven days so the engine does not undo it; negatives are claimed permanently. Ask for hands-off on an entity and the engine never touches it.
The engine finds its own targets from your search-term history, but you know things it cannot see yet — a launch keyword, a competitor's ASIN, a term that must never show your product. You feed those in through the agent.
Add keywords and product targets
Ask in plain language, including match types, competitor ASINs and categories:
- "Add 'insulated lunch bag' as exact and phrase to my lunch bag campaign"
- "Target B0XXXXXXXX and B0YYYYYYYY from my garlic press ad group"
- "Start a campaign on 'minimalist garlic press' for my garlic press"
The agent shows an Approve / Deny card with the exact request, and nothing reaches Amazon until you approve. By default it creates one campaign per keyword — the engine's own convention, so each term's performance is readable on its own. See Manage campaigns with the agent.
Negatives — usually the higher-value ask
- "Which search terms spent money in the last 30 days and never sold?"
- "Negative exact those on the campaigns they ran in"
The agent finds the bleeding terms and negates them in the same conversation, with the same approval card. A negative the agent adds is blocked permanently: the engine's own negative-keyword reconciliation will not remove it.
Why the engine does not undo your changes
Every bid, target or campaign the agent sets for you is recorded as a hold — a seven-day claim that renews while you keep working on the entity. The engine skips anything held rather than moving it back toward your target. When the hold lapses, or you release it, the engine resumes optimising that entity on its next run.
Two explicit instructions you can give on top:
- Hands off. "Never touch my Brand Defence campaign." The engine leaves that entity alone permanently, until you release it.
- A bid ceiling. "Keep bids on this keyword under $1.20 for the next 30 days." The engine stays beneath your number. A ceiling only ever lowers what the engine would bid; to bid higher and have it stick, set the bid through the agent, which records a hold.
Ask "what holds and directives are active?" to see the list, and "release the hold on…" to hand an entity back. Releasing a blocked negative lets the engine's reconciliation delete that negative from Amazon, so the agent warns you before doing it.
Two honest limits: a run already in flight when you make a change can finish with the old value, and on the rare run where the engine cannot read its claims it optimises unguarded rather than stopping. Protection is certain from the next scheduled run onward.