How it ran

1. Plot the curve before accepting the premise, rather than starting from the stated symptom.
2. Read impressions by search intent and median position, never as aggregate volume.
3. Isolate three sources of hollow impressions: brand queries where you sit behind the vendor itself, generic queries whose results page belongs to the vendors, and traffic at 85% desktop, hardly compatible with French B2B.
4. Look for the counter-example that clears the site: the homepage alone runs a 9.88% click-through rate.
5. Cross-check with GA4 filtered by hostname, without which the multi-site property inflates figures by nearly 20%.
6. Measure the winning recipe through an internal experiment: four pages, one single topic, editorial form as the only variable.

Outcome

• Impressions from 79 to 32,042 a month, clicks from 1 to 35, click-through rate from 1.27% to 0.109%.
• 1,992 impressions in positions 1 to 5 with zero clicks: the anomaly that pointed to non-human traffic.
• Recipe identified and measured: a precise, sourced question ranks at a median position of 9, against 53 for a generic page. And 53% of those pages reach the top 10, against 4% for generic ones, same domain and same week.
• Conclusion applied to content: precision of the question answered and sourced facts, not volume of prose.