Why Amazon Ads are not working: identify what is failing before changing settings
"Not working" can mean no delivery, no clicks, no purchases, no contribution after costs or reports that appear inconsistent. Those are different problems. Start with the first weak stage and a clear definition of what the campaign was supposed to achieve.
This is a troubleshooting map for authors, focused on Sponsored Products. It routes specific delivery and click issues to the no-impressions checklist and the no-click checklist, then develops the purchase, earnings and decision branches. It does not promise that one setting can make every book profitable.
| Symptom | First question | Do not assume |
|---|---|---|
| No impressions | Is the exact ad eligible, active and able to match? | The cover or conversion rate caused a delivery block |
| Impressions, no clicks | Is the exposure relevant and sufficient to interpret? | Every displayed ad was carefully noticed |
| Clicks, no purchases | Does the traffic match the offer, and are outcomes mature? | Every interested reader should buy immediately |
| Purchases, weak contribution | Does author income cover acquisition and other costs? | Retail sales equal author earnings |
| Conflicting reports | Are definitions, dates, editions and currencies aligned? | Different totals prove missing money or an account defect |
State the campaign's job
A bounded discovery test can succeed by identifying a useful reader context, even without proving recurring profit. A recurring acquisition campaign needs a stronger economic case. Define the job, allowance and next decision instead of changing the objective after every disappointing report.
1. Build a small evidence packet before editing
- Marketplace, account type, campaign and advertised ASIN.
- Edition, language, visible price, availability and applicable contribution.
- Status notices, dates, targeting, negatives, bids, modifiers and budget rules.
- Impressions, clicks, spend, purchases and sales with report scope and extraction time.
- Recent price, asset, campaign and external-promotion changes.
Amazon's book-reporting guide describes different reports for different questions. A targeting report and a click-filtered search-term report are not interchangeable. Keep a baseline export so your proposed repair can later be interpreted.
Separate observation from interpretation
Useful note: "Campaign X has 24 clicks and $12 spend in the selected period. Purchase reporting is recent. The selected targets describe beginner gardening, while the book is an advanced greenhouse manual."
That note identifies traffic mismatch and reporting uncertainty. "Amazon does not like my book" identifies neither. Do not convert a narrow observation into a claim about the entire algorithm, and do not invent first-hand campaign results to make the diagnosis sound authoritative.
Check whether the period is ready for a conversion judgment
Amazon's attribution help distinguishes account and campaign rules, including a 14-day click lookback for Sponsored Products authors. Recent interactions can gain qualifying conversions later. Compare similarly matured periods and record the extraction date.
Metrics documentation also describes reporting estimates and adjustments. Waiting for a clearer report does not require leaving a campaign active beyond its loss allowance.
2. Route the delivery and click branches without mixing them
No impressions: eligibility and configuration first
Check account/marketplace, dates, moderation, child-level status, book availability, actual limits and relevant matching. Delivering indicates eligibility, not guaranteed volume. Use the detailed delivery guide before raising bids or replacing creative.
Impressions without clicks: relevance and visible offer
Check the raw counts, selected context, placement breakdown where available, thumbnail clarity and the exact edition's offer. Empty search-term rows can be expected when there are no clicks. Follow the no-click workflow; do not manufacture missing query evidence.
A first click is useful but not the campaign's final outcome. Conversely, a tiny unclicked sample is not proof that a new cover is necessary. Move to the next branch only when the data actually describes it.
3. Investigate clicks that do not become purchases
Start with what readers were promised
As click-producing search terms and product contexts become available, compare them with the actual contents. A beginner promise, an advanced sample and an ambiguous cover can create interest followed by rejection. Inspect the destination on a phone, including edition, description, sample, price and availability in the advertised country.
| Possible mismatch | Evidence | Bounded response |
|---|---|---|
| Wrong reader intent | Clicked terms or products conflict with the book brief | Refine the relevant traffic, not every target simultaneously |
| Wrong edition | Advertised format differs from the promoted offer | Verify the intended ASIN and destination |
| Unclear promise | Description and sample do not explain the same book | Repair clarity with truthful copy |
| Price or availability issue | Offer differs from expectations in the target country | Correct the offer or stop the affected test |
| Insufficient mature activity | Small sample or recent interactions | Observe within the remaining allowance; do not declare a permanent conversion rate |
There is no universal number of clicks that guarantees a purchase. Relevant clicks can be uneconomic even if an order eventually appears. Record what you are willing to spend to learn; pausing at that limit does not require proving the book can never sell.
Do not create fake reviews, require positive ratings or alter the book to imitate an unrelated bestseller. A marketing repair should make the real book easier for suitable readers to evaluate. Book moderation guidance requires accurate creative aligned with the destination.
4. When sales exist, calculate what remains after advertising
Use the actual edition's author contribution rather than the retail-sales column. KDP's eBook royalty guidance explains why options, territories and delivery costs matter; print editions need their own calculation. State which costs have already been deducted and which overhead remains.
| Case | Spend / purchases / sales | Contribution after ads | Interpretation |
|---|---|---|---|
| A | $12 / 0 / $0 | $0 - $12 = -$12 | Check intent, destination and maturity within the allowance |
| B | $18 / 8 / $40 | $20 - $18 = $2 | Small positive contribution before other costs, not guaranteed net profit |
| C | $35 / 10 / $50 | $25 - $35 = -$10 | More purchases do not make acquisition sustainable |
Case C has ten purchases but loses $10 under the stated model. Its 70% ACOS and approximately 1.43 ROAS describe retail sales, not the author's return. The values are teaching assumptions, not actual KDP royalty rates or campaign outcomes.
Use ACOS and ROAS with contribution-based thresholds. Keep non-duplicated estimated KU income in a separate labeled model. Do not assign all account-wide reading or hoped-for future series sales to the campaign to make a loss disappear.
5. Plan one repair with an observation and a reversal rule
Write a test card
"Observed problem: [specific stage and evidence]. Hypothesis: [one plausible cause]. Change: [one meaningful variable]. Keep stable: [edition, price and other relevant settings]. Allowance: [cash/time]. Review: [condition and date]. Revert or pause if: [defined boundary]."
For a targeting problem, test a defensible reader group. For an acquisition-cost problem, inspect adjusted bid exposure. For a cash-authorization problem, use budget controls. Do not raise bids and budgets together simply because sales feel slow.
Confirm the new asset or setting is actually in effect, save the before/after configuration and avoid a second edit before the first can be understood. A normal before/after comparison is not a randomized experiment; log discounts, seasonality, emails and other changes that could affect results.
6. Know what support can investigate and what it cannot promise
KDP's advertising help distinguishes initial setup guidance from technical, eligibility, billing and policy investigations handled by Amazon Ads support. Escalate inconsistent status, catalog identity, unexplained billing or technical errors through the official channel with the evidence packet.
A valid support case is specific: account/marketplace, campaign ID, ASIN, dates, exact notice, matching report scope and checks already completed. Do not provide passwords, SSH keys, full payment-card details or unrelated personal records. Retain the case ID and the explanation so future checks do not restart from zero.
Support can clarify or investigate a block; it cannot guarantee reader demand, a profitable CPC or a successful book. If the campaign is eligible but the offer does not persuade suitable readers at an affordable cost, that is a publishing and marketing decision, not automatically a software defect.
7. Choose continue, redesign, pause or stop
- Continue only with a clear question and an affordable remaining allowance.
- Redesign when evidence identifies a fixable reader, offer or configuration mismatch.
- Pause while resolving a broken destination, uncertain status or cash constraint.
- Stop when further learning is not worth its cost or the book cannot support acquisition.
- Reconcile matured reports after pausing; do not erase the original observation.
Stopping ads does not mean abandoning the book. A sample, a consenting-reader newsletter or a relevant permitted promotion can test different discovery routes. Use the starter marketing guide and the no-paid-distribution plan when advertising is not the right current channel.
Return to delivery diagnosis, click diagnosis or the Ads hub according to the stage your evidence supports.
Sources and scope
Official Amazon Ads and KDP documentation checked: 14 October 2026. Test cards, campaign cases and dollar amounts are fictional editorial examples. They are not an audit of your account, a universal minimum data requirement or a promise of recovered sales. Attributed results do not by themselves establish incremental profit.