Applying one click-through curve to every query overstates traffic by 40%
A single CTR-by-position model (position 1 equals 30%, and so on) ignores that the curve bends with intent. Across 2,000 ranked queries:
— Navigational, position 1 actual ▇▇▇▇▇ 52%
— Informational, position 1 ▇▇▇ 28%
— Long-tail question, position 1 ▇▇▇▇ 41%
— Forecast error from using a flat curve ▇▇▇ 39%
Navigational intent concentrates clicks at the top; broad info spreads them down the page.
The mistake: one traffic model for all keyword types.
The fix: bucket terms by intent first, then apply a separate click-through curve per bucket.
Benchmark of the week: a flat CTR model overstates informational-term traffic by about 39%.

Your SERP screenshot is personalized, so your difficulty read is wrong
Checking difficulty from your own logged-in browser bakes in history, location, and engagement. Across 300 queries pulled both ways:
— Personalized vs clean (incognito, neutral geo) result overlap ▇▇▇ 64%
— Own site ranking higher in personalized view ▇▇▇▇ 38% of cases
— Average position inflation ▇▇ 3.2 spots
— Terms judged 'already ranking' that were not ▇▇ 21%
You see yourself ranked, you skip the term, you leave demand on the table.
The mistake: assessing competition from a personalized result page.
The fix: pull SERPs incognito with a set location and no login, or use a neutral API, before scoring difficulty.
Benchmark of the week: personalized views inflated our own position by about 3.2 spots on a third of terms.