The word you choose can cost you 82 points of difficulty

Every figure below was read in Semrush’s United States database on 25 and 26 September 2026, one keyword per lookup. The left-hand term names a discipline — the field of work. The right-hand term names a deliverable or a platform — the thing you are hiring someone to do, or the system you need it done on.

Six pairs, six gaps, none narrower than 33 points. The widest is 82: “schema markup” measures 97% difficulty and “schema markup service” measures 15%. One word between them.

A discipline term is what you search when you are trying to understand something. “Schema markup” is typed by developers looking for the documentation. “Technical SEO” is typed by marketers reading up on their own craft. “Generative engine optimization” is typed by consultants deciding whether to add it to a menu. Semrush classifies all six discipline terms above as informational.

The results page reflects that. Definition articles, vendor explainers and the official documentation compete for those queries, published by organisations with a decade of accumulated authority. Schema.org and Google’s own structured-data guidelines are not pages you outrank with a service page.

A deliverable term is what you search when you have already decided. Nobody types “shopify seo agency” to find out what Shopify is. Nobody types “schema markup service” to learn what structured data means — they know, and they have decided they do not want to implement it themselves. The competing set is much smaller, because it is only firms who have built a page for that specific job.

Cost per click is what advertisers will pay for a visitor, which makes it the closest public proxy for commercial value. We assumed it would track difficulty — harder term, bigger prize. It runs the other way, in every pair we measured:

“Schema markup” is 82 points harder than “schema markup service” and worth roughly a third as much per visitor. “Technical SEO” is 43 points harder and worth less than a third. The pattern repeats in both AI-search pairs.

This is the part worth sitting with. The harder phrase is not a bigger prize being guarded. It is a smaller prize that costs more to reach. Difficulty and value are not two ends of the same trade-off here — they point in opposite directions, because difficulty is set by how many people write about a subject and value is set by how many people want to buy it.

It shows in the general terms too. “Seo services” and “seo agency” both measure 60,500 searches a month in the US, but “seo services” sits at 60% difficulty and $14.37, while “seo agency” sits at 41% and $17.54. Same volume, 19 points apart, and the easier one pays better.

We expected these to be the open ground. The category is barely two years old, and the assumption inside our own plan was that arriving early was an advantage.

It is not, at the head. “Generative engine optimization” measures 77% difficulty in the US — harder than “seo services” at 60%, harder than “technical seo” at 62%, harder than “ecommerce seo agency” at 52%. In under two years the largest agencies and the SEO software companies have taken the definitional ground completely.

The hiring variants are untouched: 25% and 24%. Whatever early-mover advantage existed in AI search has already left the head terms. It survives, for now, in the terms people use when they are buying.

Our first pass at this had six pairs and was wrong.

Two of them — the technical SEO and schema markup rows — compared a discipline-side figure measured in the India database against a deliverable-side figure measured in the US database. Difficulty is scored against the competing set inside one country database. Comparing across two of them is not a finding, it is a category error, and we published the claim internally before catching it.

The fix was to measure both terms again in the US database, which is what the table above now contains. As it happens the corrected figures are stronger than the ones we had: schema markup read 79% in India and 97% in the US, which turned a 64-point gap into an 82-point one. Being wrong in our favour is still being wrong.

We are telling you this because the mechanism that caused it is probably in your spreadsheet too. Most keyword research is kept in sheets with columns for volume and difficulty and no column for the market the figure came from. Once two exports from two databases are pasted into the same sheet, every comparison drawn from it is unsafe, and there is nothing in the data to tell you which rows are the problem. Ours now has a database column. That is the whole fix.

• Record the database with every figure. If your keyword sheet has no market column, half your comparisons may be unsafe and you cannot tell which half.
• Name the deliverable, not the field. “Technical SEO audit service” rather than “technical SEO”. “Schema markup service” rather than “schema markup”. The buyer is the same person; the door is not.
• Check the CPC before assuming the big term is the prize. In all six pairs here, the low-difficulty term was worth more per visitor than the high-difficulty one beside it.
• Let the platform qualify the visitor. “Saas seo agency” at 8% difficulty is not a small ambition. It is 3,600 US searches a month from people who have already described their own business in the query.
• Do not read low volume as low value. “Schema markup service” draws 170 searches a month. They are 170 people who have decided to pay someone. The 8,100 on the head term are mostly developers looking for documentation.

None of this makes a head term unwinnable. It makes it a multi-year project rather than a quarterly one, and it says that a site without an established link profile should spend its first year in the right-hand column.

This site now practises it. On 26 September 2026 eight of our service pages were retitled from the discipline to the measured deliverable: SaaS SEO Services became SaaS SEO Agency, Technical SEO Services & Audits became Technical SEO Audit Service, Schema Markup & Structured Data became Schema Markup Service, and so on down the list.

We have no results from that yet and will not have for months. When there are, they will be published here with their dates — including if they show nothing. That is the arrangement on this blog: the working, not just the conclusion.