Stop Publishing. More Than Half My Clients' Impressions Were Already Sitting in Positions 4 to 20.

Publishing more keyword pages has stopped moving the needle. Architecture is what moves it.

The loudest advice in SEO this month is the opposite of the advice from three years ago. Stop writing. Stop feeding the calendar a new page for every keyword variant. Fix what you already have: the internal links, the folder structure, the near-duplicate pages fighting each other, the pages that already sit on page one and two and never get clicked.

There is now a number behind the topical half of that argument. Semrush tracked 1,094 categories in ChatGPT from January through June and published the split: in categories close to a brand's core expertise, the brand was both cited and named in 34 percent of appearances. In categories far from it, 9 percent. Depth in a few topics beats a thin presence in many.

That is the AI half. I wanted the plain search half, for the people I actually work for, a dentist and an HVAC company and a window installer, not a SaaS brand with a content team. So I opened their Search Console data and asked one question. Where are the impressions sitting right now, before anybody writes another word.

2,500 queries, five businesses, 28 days, sorted by position

The set: the five Local Howl clients with Search Console connected, a dentist in Florida, an HVAC company in Alabama, a window company in Missouri, a vehicle wrap shop in Texas, and a pool builder in Georgia. The window: August 31 to September 28. The data: each client's top 500 queries by Search Console's own ranking, 2,500 queries total, bucketed by average position.

Positions 1 to 3: 250 queries, 4,275 impressions, 259 clicks. That is 9 percent of the impressions and 64 percent of the clicks. Click-through rate, 6.06 percent.

Positions 4 to 10: 540 queries, 19,396 impressions, 116 clicks. 40 percent of the impressions, 29 percent of the clicks. Click-through rate, 0.60 percent. Ten times worse than the top three, on the same first page.

Positions 11 to 20: 416 queries, 6,986 impressions, 24 clicks. Click-through rate, 0.34 percent. Positions 21 and beyond: 1,294 queries, 17,959 impressions, 8 clicks.

Put 4 through 20 together and it is 956 queries holding 54 percent of all the impressions in the set, earning 140 clicks. If those same impressions earned the top-three click rate, that would be about 1,600 clicks instead of 140. That is arithmetic, not a promise, but it is the size of the prize sitting on pages that already exist.

The page view tells the same story louder. Of the 125 top pages across the five sites, only 7 average a top-three position, and together they had 64 impressions. 73 pages average somewhere between 4 and 20, and they carry 81 percent of all page impressions and 91 percent of all page clicks. The sites are not short on pages. They are short on pages that finish the climb.

The inventory is already built. It is parked one shelf too low.

The dentist is the cleanest example. 297 of its queries sit in positions 4 to 20 with 16,168 impressions and 70 clicks. Its 31 top-three queries had 811 impressions and 34 clicks. One single query averaging position 5.8 showed up 3,541 times and got clicked 9 times. Nobody needs to write that dentist a new blog post. Somebody needs to get the page that already ranks fifth into the top three.

The HVAC company has a query averaging position 5.9 with 892 impressions and zero clicks. The pool builder has 89 queries in positions 4 to 20, 921 impressions, and zero clicks. Page one, no phone calls. That is not a content gap. It is an architecture gap: the page is relevant enough to show and not trusted enough to win.

One honest exception. The window company does its work at the top: its top-three queries had 1,048 impressions and 144 clicks, and its 4 to 20 band had 786 impressions and 2 clicks. Its next win is not this playbook, it is the map pack and new demand. Run the numbers before you pick the strategy. The data picks it for you.

This also lines up with what I found last week when I tested whether AI cites boring SEO. The fundamentals did the work there too. The engines rewarded the businesses that were already organized and consistent, not the ones with the most pages.

Restructure before you write. Four moves, in this order.

One. Pull the band. Export Search Console queries for the last 28 days, keep positions 4 to 20, sort by impressions. That list is your content calendar for the quarter, and every item on it already has a page.

Two. Point your strongest pages at the ones on the edge. Find which page ranks for each top query, then add internal links to it from the pages that already get clicks, with anchor text in the words people actually searched. This is the cheapest ranking signal you control, and almost nobody uses it on purpose.

Three. Merge the twins. Local sites are full of near-duplicate city and service pages splitting the same query between them. Pick the stronger one, fold the other into it, redirect. Two pages at position 9 are worth less than one page at position 3.

Four. Own fewer topics, on purpose. The 34 percent versus 9 percent split says the same thing for AI answers that the click data says for search: depth in a core category gets you named, scattered coverage does not. Decide which three services you want to be known for and build the links and the pages around them. Then measure what changes, which is harder than it sounds, because Search Console hides a big slice of the clicks.

Find the band, fix the links, check the depth

Search Console tells you which queries are stuck. It does not tell you why. The SEO Toolkit covers the why: Site Audit shows the internal linking structure, the orphaned pages, and the pages with almost no internal links pointing at them, which is usually where the position-7 pages are hiding. Position Tracking watches the band move week to week, and Competitive Research shows which topics the businesses outranking you actually cover in depth, so you know where to consolidate and where you are genuinely thin.

Then check whether the depth is paying off in the AI answers too. The AI Visibility features in Semrush One track whether a brand is cited and named across ChatGPT, Perplexity, Gemini, and Google AI Mode, which is the payoff metric behind that 34 versus 9 split. Same restructuring work, two scoreboards.

You do not have a content problem. You have a shelf problem.

Five businesses is a small sample, and position is an average, so treat the buckets as a map, not a law. But four of the five told the same story. More than half the impressions were already on page one and two, collecting a fraction of the clicks. The cheapest growth in local SEO right now is not the next page. It is the last three positions on the pages you already paid for.