Volume is a smoothed estimate and impressions are a count, and the difference is not a rounding error. Southwell Saddlery, which repairs and fits saddles and bridles in Nottinghamshire, took six queries it already appeared for, wrote down what two keyword tools estimated for each, and then read what Search Console had actually recorded over the twelve months to June 2026.
The six read as follows, in monthly figures: saddle fitting nottinghamshire estimated at 90 and 140, actual 260. Bridle repair near me was estimated at 70 and 210 and delivered 540. Saddle flocking cost, estimated at 30 and 40, actual 110. How to measure a horse for a saddle, estimated at 720 and 480, actual 190. Second hand saddles, estimated at 1,900 and 2,400, actual 60. Wool flocking versus foam, estimated at 20 and 10, actual 95.
What the six say
The error runs both ways, so a safety margin in one direction is wrong. Three queries delivered several times what either tool predicted and two delivered a fraction of it, and the third pattern is the one that matters commercially: the two biggest estimates, second hand saddles and how to measure a horse, are the two that under-delivered most.
The two numbers
What two tools estimated, and what was actually recorded
The two estimates disagree with each other on four of the six rows, and the largest estimates are the ones furthest from what this shop ever saw.
Monthly figures for one saddler, twelve months to June 2026. The third column is a count, the first two are estimates.
The third column counts impressions for one site, so a national query with a local answer looks tiny here and is not. The tools are not wrong about the query; they are answering a different question.
Both of those are national queries with local answers. A saddler in Nottinghamshire appears for second hand saddles almost nowhere, so the 1,900 people typing it are real and are not this shop's impressions. The tool is not wrong about the query. It is answering a different question from the one the shop is asking.
The two tools disagreed with each other more than either disagreed with reality on four of the six, which is worth sitting with before treating either as a measurement. On bridle repair near me they gave 70 and 210, a threefold spread, and neither was within a factor of two of the 540.
What difficulty actually measures
Difficulty is a statement about the pages ranking now, not about yours. The score is built mostly from how many other sites link to whatever currently occupies the first page, which is a fact about the past and about other people. It contains nothing about the page you are going to write.
Difficulty scores
What a difficulty score sees vs. what decides if you can win
A difficulty number summarizes links pointing to today's ranking pages and stays blind to whether those pages actually answer the query.
- Links to current ranking pagesReflects past authority earned by other sites, not your future draft.
- Identical easy scores across unlike queriesGives wool flocking versus foam the same grade as three queries the shop cannot touch.
- Wrong page format in top spotsThree product listings ranking for a comparison query signal an opening.
- Unanswered depth on the topicNobody had written properly about wool flocking versus foam yet.
- Strength of the page you write nextNo link-based score measures content fit before publication.
Reading the actual top results shows format gaps that link-count scores cannot detect.
The practical consequence at Southwell was wool flocking versus foam, scored as easy by both tools, and it is easy: nobody has written properly about it. It was also scored the same as three other queries the shop cannot touch. A difficulty score cannot see that the top three are all wrong for the query, and on that one they were three product listings answering a comparison question, which is the single best signal there is that a page could win.
Where this stalls: the sorted spreadsheet
The stall is a list of four hundred rows sorted by volume, worked from the top. Every number in that column is an estimate of a different quality, the largest ones are the least reliable for a small local business, and sorting puts exactly those at the top.
Sorting workflow
Sort keyword lists into three decision piles, not a volume ranking
Sorting four hundred rows by volume puts the biggest national mismatches at the top; order-of-magnitude piles prevent wasted work.
- Worth a dedicated page2
- New pageQueries calibrated against own impressions
- PrioritizeSpecific local or comparison questions
- Worth a paragraph2
- Add sectionSecondary details inside an existing service
- MergeSub-topics too narrow to stand alone
- Belongs to somebody else2
- SkipHigh-volume national terms with local mismatch
- DropUnverified top-sorted spreadsheet traps
One hour checking estimates against live Search Console impressions changes how every future row is sorted.
Use volume to sort into piles, never to forecast. Three piles is enough: rows worth a page, rows worth a paragraph on an existing page, and rows that belong to somebody else. None of those three decisions needs a number more precise than an order of magnitude, which is fortunate, because that is all the number is.
Check one estimate against your own impressions before trusting a list of four hundred. The exercise above took an hour and it changed how the shop reads every list it will ever build, which is a better return than any single row on the list.
Where the estimates are worst
Seasonal queries are where a smoothed average is worst. Saddle fitting is a spring and autumn trade at Southwell, and the annual average conceals a March that runs at four times a January. A tool reporting one number a month for a query with that shape is right about the year and wrong about every month in it.
Where estimates break
Three reasons monthly keyword estimates drift from real demand
Annual smoothing, spend-gated tool ranges, and real Search Console impressions behave differently when demand swings by season.
- Seasonal averages hide peaksSaddle fitting runs four times higher in March than January, yet tools report one flat monthly figure.
- Keyword Planner hides behind rangesFree accounts without active ad spend receive wide bands instead of single numbers.
- Search Console logs real impressionsFree since replacing analytics keyword data in 2011, counting actual appearances for your site.
Where a smoothed estimate and Search Console disagree, Search Console records what happened rather than an average.
Google Keyword Planner supplied one column, free with a Google Ads account, and it reports wide ranges rather than figures when the account is not spending, which is more honest than it looks. Search Console supplied the actual impressions, free, and it replaced the keyword data Google removed from analytics in 2011. Where the two disagree, the second one is not an opinion.
What the six queries actually want, which decides whether an impression could ever become a visit, is what the searcher wants, which the words hide. The queries with the smallest estimates and the best behaviour are specific queries, less competition, better conversion, and where the list came from before any number was attached is sources, and building the first list. All of it is finding what people type, and what they mean by it with two unreliable columns bolted on.