Workflow: Auction pressure attribution
Before explaining a rise, establish that there is one in the market data at all — and when it started.
Add your data
Adding details ensures a relevant result
Save time by automatically applying details and getting ready-to-use prompts.
Prompt
For {your-domain.com} in {country}, return the last 12 months of monthly paid totals.
Return these columns:
• month
• paid_keywords
• estimated_paid_traffic
• estimated_paid_cost
• implied_avg_cpc
Calculate implied_avg_cpc as estimated_paid_cost divided by estimated_paid_traffic, rounded to two decimals. Label it no_data where either input is missing or zero.
Then state below the table:
• inflection_month — the month after which implied_avg_cpc rises and does not return to its earlier level
• the percentage change in implied_avg_cpc from the first month of the series to the most recent
• the percentage change in paid_keywords over the same span
If implied_avg_cpc shows no sustained rise, do not stop. State plainly that the portfolio-wide average does not show the increase, and that this points either at account-side causes such as bidding strategy, quality score or conversion rate, or at a mix shift — a flat average can hide sharp rises on individual keywords, especially where the paid keyword count contracted over the same period. Then set inflection_month to the month with the largest single-month rise in implied_avg_cpc, label it no_market_inflection, and continue to Step 2, which tests the mix-shift explanation keyword by keyword.
If {your-cpc-change} was provided, compare it to the market-side change and classify the gap in one sentence:
• market_driven — the two are within 10 percentage points
• account_driven — yours materially exceeds the market
• better_than_market — yours is lower
Is this prompt helpful?
Example Output (illustrative)
- A table of 12 rows — one month per row with paid keyword count, estimated paid traffic and cost, and a derived implied average CPC
- A short block underneath naming the inflection month, the CPC and keyword-count change across the series, and — where your own figure was supplied — a one-sentence market-driven or account-driven verdict
- Or, where no sustained rise exists, an explicit statement that the market-wide average does not explain the increase, a no_market_inflection label, and the month of the largest single-month rise carried forward so the remaining steps still run
