6,482 Page-One Rankings, 218 Clicks: Five Things We Fixed

Our own Search Console says margen.net holds 6,482 page-one rankings. Those rankings produced 218 clicks in 90 days.

I’m publishing it because it’s the most useful number we have, and because almost nobody shows you theirs. Below is what 90 days of our own data actually said, the five things we changed, and the two things we tried that did nothing. Every figure is from margen.net, pulled the week of writing.

None of it required publishing a new page.

1. Check your metadata is broken before you try to optimise it

The usual advice is to rewrite titles and descriptions for click-through. Fine. But run a check for plain breakage first, because we found ours had some.

Thirteen live pages were rendering raw HTML entities in the browser tab and in search results. Titles imported from WordPress years ago still carried – and ’, and the template engine escaped them again on the way out, so what appeared on screen was the literal text – rather than a dash. Nobody had noticed, because nobody looks at their own tab titles.

The second one was duller and bigger. 186 of our 422 pages had a meta description over 160 characters. Anything past roughly that point gets cut off in the result, so the end of the sentence never appears. Our longest was 267 characters. One of them, on a page absorbing thousands of impressions, was severed mid-word at “never done befo”.

How to check both in an afternoon: crawl your own site, export every title and description, count the characters, and search the output for &# and for any character your house style bans. We found 82 em dashes sitting in live titles and descriptions on a site whose style guide forbids them.

Fix the breakage first. A truncated description is a measurement problem before it’s a writing one, because you can’t tell whether your words work when half of them never render.

2. Impressions without clicks is often not a click-through problem

This is where the received wisdom let us down hardest.

The standard move is to find pages with high impressions and low click-through, then rewrite the title to pull people in. We did the analysis. Here is what it showed.

6,442 of our 6,482 page-one queries got zero clicks. Not low click-through. Zero. Between them they carried 65,650 impressions.

Before anyone concludes our titles are terrible, we split those queries by whether a person or a machine was doing the asking:

So roughly 59% of it is assistants fanning out queries and reading the answer without anyone clicking anything. That’s the channel working exactly as we tell clients it works, and it’s invisible in a click report.

The other 26,991 impressions are people. Still zero clicks. Those are worth looking at individually, and when we did, the top of the list was not a title problem either:

• copilot seo agency, 671 impressions, position 6.8, no clicks
• how to fix chatgpt misinformation about my brand, 421 impressions, position 1.4, no clicks
• best geo tool in great britain, 328 impressions, position 4.7, no clicks
• is paying someone to do seo worth it, 309 impressions, position 8.2, no clicks

Position 1.4 with no clicks isn’t a headline you can rewrite your way out of. Something is answering above you.

The finding that changed how we pick targets

While we were in there we banded every query by how often it was searched. The pattern was clean and the opposite of what most people assume.

The more people search a term, the worse we rank for it. Volume and competition move together, and at our authority the broad terms belong to publishers many times our size.

I’m a strong believer that this is the most useful thing a smaller site can internalise. The instinct is to chase the big number in the keyword tool. The data says the big number is the one you are least likely to win, and the specific question is where you actually place.

3. Count internal links in the body, not on the page

Everyone says build internal links. Almost nobody says where to put them, and that turned out to be our problem.

Our ten statistics pages are the best-performing thing on the site. They average position 8.2 while the rest of the corpus sits between 21 and 59. So we checked how well they routed traffic onward, expecting to find they were fine.

All ten had zero internal links in the body. Every single link sat in the closing block after the last section. A reader who read two thirds of a 1,200-word page and stopped had no route anywhere at all. Five of the ten had no commercial link whatsoever.

The second check was inbound rather than outbound. Our AI SEO service page had 15 inbound internal links. Its sibling GEO hub had 32. Three pages in the same topical cluster, all about pricing and scope for that exact service, linked to it zero times.

Two counts worth running on your own site, and they take minutes:

1. For your best pages, how many links sit above the final section?
2. For your money pages, how many inbound internal links, and which sibling pages ought to link and do not?

On anchors, the advice in the post that prompted this is right and worth repeating: use the terms Search Console already shows you ranking for. You are telling the engine the two pages are about the same thing, in language it has already associated with you.

4. Rebuild the thin page that is outranking your conversion page

Refreshing content is the oldest advice in SEO. The version worth doing is narrower than “update your old posts”.

Find the page that is absorbing demand meant for something else.

Ours was a 611-word blog post about auditing AI visibility. It was taking 2,384 impressions of audit demand at position 36 to 46. The free audit tool it should have been feeding, the actual conversion page, was getting 52.

A thin post was outranking our own tool on the tool’s own topic, and converting nobody.

The temptation is to redirect it. We didn’t, because it holds a position we’d be throwing away, and because the method in it was sound. We rebuilt it instead: 611 words to 2,108, the free tool moved into the first sentence rather than the closing block, eight questions added with the schema to match, internal links from four to eight.

The rule we took from it: if the thin page ranks, rebuild it. If it does not rank, redirect it. Don’t redirect away a position you already hold just because the page embarrasses you.

5. Before adding topical depth, check you are not competing with yourself

Topical authority is real. Adding pages is how most people chase it, and on a corpus of any size that is how you quietly start ranking against yourself.

We mined 90 days of Search Console: 13,202 distinct queries, of which 4,678 were question-shaped. Then we checked how many already had a page of ours ranking in the top 20.

3,326 of them did.

On a 328-post site, most questions are already answered by something we published. Building a page for those would not add authority, it would split it.

So the check runs before the writing, not after. Pull the query, look for an existing page of yours in the top 20, and if one exists the job is improving that page rather than writing a new one. It’s slower and less satisfying than a content calendar. It’s also the difference between a corpus and a pile.

The two that did not work

Publishing only the wins makes the rest untrustworthy, so here are the two that failed.

The inline audit widget. We put an interactive AI Visibility Score widget directly into five high-traffic pages, on the reasoning that the tool was too far from the reader. Those five pages took 214 sessions and produced zero form submissions. Placement was not the constraint, so we stopped that line of work rather than adding a sixth.

The booking link. 376 pages on margen.net carry a link to our booking calendar. Across 28 days and 3,385 page views, it was clicked zero times. We checked the tracking against Google’s own outbound click data rather than assuming, and the tracking is fine. Nobody is clicking it.

Both cost us something to learn. Both are more useful than another list of tactics that worked on somebody else’s site.

What to do with this

Run the five checks on your own data. They need no new content and no tools beyond Search Console and a crawler:

1. Export every title and description. Count characters, search for stray entities.
2. Pull your page-one queries and count how many earn zero clicks. Separate the conversational ones from the typed ones before drawing a conclusion.
3. Band your queries by impressions and look at average position per band. Find out whether volume helps or hurts you.
4. Count in-body links on your best pages, and inbound internal links on your money pages.
5. For the last ten things you were about to write, check whether you already rank top 20 for them.

If the second one shows what ours did, the conversation changes from click-through rate to whether you are the source being quoted when the answer appears without a click. That is what we do, and the AI Visibility Score is the free version of the first check.

We publish the rest of our own numbers as we go. The method behind all of it is the Synaptic Authority Engine, and the zero-click statistics page has the wider market picture alongside ours.