360 Pages, 48 Clicks: An SEO Case Study of Four Sites That Failed

TL;DR: four sites we own, 360 pages between them, 33,615 impressions and 48 clicks across 90 days. A click-through rate of 0.14%. The sharpest comparison sits inside the site with the best content. Its three evidence pages (five timed withdrawals, four physical handsets, a fifteen-platform table) take 27% of that site's clicks from 3.4% of its URLs, so first-hand data does beat the average. But the single page with the most clicks has no first-hand data at all: a comparison piece ranking 7.6 with 191 impressions and 7 clicks, against the payments review ranking 34.9 with 14,580 impressions and 2 clicks. Seventy-six times the impressions, three and a half times fewer clicks. The conclusion is not that content does not matter. It is the inverted one. On a domain with no authority, content quality is not the binding constraint.

All four are our own assets. Sites are unnamed. Every figure comes from Google Search Console for 8 June to 5 September 2026, type: web, dataState: final, pulled on 9 September 2026 — the last section explains why that date is stated so precisely.

The ledger

We are publishing this because a cross-section like it is only available to somebody who actually did it. Most writing about how long SEO↗ takes comes from people who have not run four new sites simultaneously and then measured all four with one instrument on one day. We did, and every section below comes out of that measurement.

Method, so you can reproduce it

The numbers above are deliberately boring to obtain, and that is the point:

• Source: Google Search Console Search Analytics API, type: web, dataState: final. Note that average position in Search Console↗ is the highest position for the query, weighted by impressions, so a site average of 32 means it consistently sits that deep rather than occasionally. Not a third-party rank tracker: those are estimates of a sample, and after Google removed the 100-results parameter their coverage got worse in ways that do not show up as gaps in the chart.
• Window: one fixed 90-day range for all four sites, so seasonality applies equally.
• Page-level and query-level pulls, not just site totals. Site totals hide the two facts that mattered most here: that 57% of one site's impressions sit on a single page, and that 48 of 54 URLs on another have never had a single impression.

If you want to run the same analysis on your own portfolio, the whole thing is three API calls per site and an afternoon. The hard part is not the data. The hard part is agreeing in advance what number would count as failure.

The expensive lesson: the content was real, the site still died

Three of these sites can be explained with "the content was not good enough." Site A cannot.

Site A has first-hand assets specific enough to be checked:

• A withdrawal review: five real withdrawals, 14 January to 8 March 2026, AU$640 total, each one timed with a stopwatch, with the conditions that triggered KYC recorded.
• A mobile review: four physical handsets (iPhone 15 Pro, iPhone SE, Pixel 8, Galaxy A24), the package IDs of counterfeit APKs listed, 50 games sampled.
• A payments review: a fifteen-platform comparison table, with one platform explicitly marked as not personally tested.

Those three take 9 of the site's 33 clicks (27%) from 3 of its 88 URLs with impressions. First-hand data does beat the average.

The question is what "beats the average" is worth here. Site totals: 31,884 impressions, 33 clicks, average position 31.4. And the page with the most clicks is none of those three. It is a comparison piece with no first-hand data in it:

Read the first and last rows together: seventy-six times the impressions, three and a half times fewer clicks. The comparison page proves nothing and happens to rank 7.6. The payments table cost fifteen platforms' worth of testing and ranks 34.9. That one page alone is 45% of the site's impressions.

Position 31.4 means pages three through five. Click-through at that depth is effectively zero regardless of how good your title is, because nobody scrolls that far to read it. Site A's three highest-impression queries drew 1,160, 804 and 643 impressions and zero clicks between them.

On a domain with no authority, content quality is not the binding constraint. Authority is.

The same site has a sharper example. It has a full review targeting a competitive head term, with a testing protocol most publishers would not bother with. Tested 3 February to 10 May 2026, KYC completed with a driver's licence, withdrawals from three banks timed at 19, 31 and 47 hours.

For that head term, the page never entered the top 100. Zero impressions. It picks up one long-tail query at position 49.

Right content, wrong keyword. A zero-authority domain aiming at a head term is putting hard-won first-hand data somewhere nobody can see it. The cost is not the page. It is every hour spent producing the evidence on it.

Our threshold now: a new domain only targets keywords at difficulty 10 or below, and never head terms. Ahrefs' own explanation of how difficulty is calculated↗ is worth reading precisely because it makes clear the score is an estimate from their link graph. Whether your site can reach a given SERP is something only your own impression data can answer.

Fabricated first-hand buys you nothing

Site B's "first-hand" claims were more specific than site A's real ones:

"I tested six apps with ₹500 each (24 April – 8 May 2026, Samsung Galaxy A54)" "an 11-day testing diary" · "4 withdrawal tests" "read more than 600 reader emails" · "a notebook of 280 reader emails… 92%" "18 first-hand deposit tests… 72% success rate"

There is not a single screenshot, log or CSV in the repository.

The worst part: the site already knew. Its content config carries a comment dated 12 June 2026 saying, in substance, that invented expert personas are an E-E-A-T liability and that one of them had collided with a real person.

The comment sat there. Production continued.

Search Console settled it. The site's second-highest-impression query over 90 days is a quoted personal name: our invented author, followed by the credentials we gave him, 28 impressions and 0 clicks. The next is a quoted article headline: 32 impressions, 0 clicks.

Quoted-name and quoted-headline searches have one obvious intent: somebody is checking whether this person and this article are real. The answer was no.

Fabrication bought no rankings. It bought a liability that generates its own audit trail.

A fake author persona is worse than no author persona.

The control group was next door. Site C did the opposite, closing every article with a line saying the piece was built from published and reported sources rather than personal testing. Site C: 12 clicks. Site B: 1.

Both bases are tiny and that ratio is not a statistical result. But the direction is clear, and the important half is that honesty cost site C nothing. Saying "we did not test this ourselves" did not weaken its other claims. It is what made them credible. This is exactly the question Google's guidance on helpful, reliable, people-first content↗ asks: does this page demonstrate first-hand expertise? If it does not, do not pretend it does.

Cloning one template four times is one bet placed four times

Laid side by side, all four sites have the same shape:

Batch-clustering checks were red on all four: within each site, the articles resemble each other closely enough for a machine to identify the mould.

The real problem is not inside any one site. It is across them:

By the time site #2 launched, site #1's data was already sufficient to reject the template. Sites #3 and #4 went into production knowing the approach had not produced clicks.

Launching new sites is not diversification. It is repeating the same bet.

Diversification requires the bets to be uncorrelated. Four sites sharing one template, one content pipeline, one keyword strategy and one author system are perfectly correlated.

Our rule now: no template gets cloned to the next site until the previous one has produced clicks. One pass condition, measured at 12 weeks.

The topics were right. The sizing was wrong.

The tempting wrong conclusion is that these sites picked bad topics. They did not.

• Site C's terms are real human queries: a payment method's withdrawal process, a new piece of legislation by name and by full title (positions 10.5 and 16.8).
• Site D's terms are genuine local demand: two regional betting brands' withdrawals, bank-name mismatch resolution, and those are exactly the pages that got impressions.
• Site A's three largest queries drew 1,160, 804 and 643 impressions. The demand exists; none of it converted, because it all happened past page three.

The problem was never demand. It was whether this domain could reach it.

Site D makes it unmissable: 6 of 54 URLs have ever had an impression (re-verified 9 September, unchanged). The other 48 do not exist as far as Google is concerned.

So the question at topic-selection time has to change. Not:

"Does anybody search this?"

But:

"At my current authority, can I reach page one for this?"

The first question is almost always yes, which means it is not a decision. The second one eliminates most of your list, which is exactly what makes it useful.

Position 10 with zero clicks is a format problem

Site C has an explainer about a piece of legislation sitting at average position 10 with zero clicks. First page, no clicks at all. Site A has a "what is RTP" definition page at position 25.9.

What they share is that their answers fit inside a summary. The user gets what they came for on the results page and has no reason to click.

That is a format problem, not a ranking problem. Pure definitional content ("what is X") now feeds summaries — it builds topical relevance and it will not send you traffic. It can exist; it cannot be the main line.

The main line has to be operational: how-to content and tools. Things a user has to follow or use, which a summary cannot take away because the user needs the whole sequence or a calculation.

Claiming verification without linking is self-defeating

Site D has zero external links site-wide, and every article says every figure was checked against published policy.

Site B has 88 articles citing fewer than three domains, 67 of them citing none. Site C has six at zero.

Claiming verification without sources is worse than not claiming it.

Not claiming is merely no credit. Claiming and not linking actively converts "this is checkable" into "this is visibly uncheckable" — you invite the reader to verify and then withhold where.

That became a hard rule: if you write "verified", you link. If you cannot link, do not write it.

Two facts that contradict the site itself

Two of site C's articles have problems more urgent than any score:

One: recommending services that stopped offering the thing. A "real-money games" roundup lists ten apps as paying real money in 2026. Those apps all ended cash play after an August 2025 regulatory change, and the site's own tracker article says so. The site contradicts itself.

Two: the wrong brand name on the about page. The "how we review" page is headed with the previous site's brand, copy-pasted from v1 into v2.

These matter more than sixty articles scoring below 80. A low-scoring article is merely unhelpful. A self-contradicting site is destroying the only asset it has, and adding sources to it is underwriting an error.

An asset without version control is not an asset

Site D has 221 files that were never committed, and its git HEAD still points at another site's commit. A fifth site of ours, outside this group, is not a git repository at all.

This is asset management rather than SEO, but it becomes SEO immediately:

• You cannot roll back a bad change
• You cannot tell which version is live
• You cannot hand it to anybody else

A site without version control is a site whose current state you do not know, and you cannot optimise what you cannot describe.

How we would size the next one

Everything above turns into nine rules, and none of them are general advice:

1. Compute reach before launch. Domain authority sets your maximum keyword difficulty. Difficulty ≤ 10 for a new domain, no head terms.
2. Fix what exists before starting something new. Moving a site that already has 31,884 impressions from position 31 to page one costs far less than 12–18 months from zero.
3. Never fabricate first-hand. If you did not test it, say "built from published sources, not personal testing." Honesty is free; fabrication is a liability.
4. No cloning until the previous template produced clicks. Measured at 12 weeks.
5. Select topics on "can I reach page one", not "does anybody search this".
6. Definitional content is not the main line. Operational how-to and tools are.
7. Claim verification only with links.
8. git init on day one. No version control, no launch.
9. Define the kill condition before launch. Fewer than N clicks in 12 weeks means stop adding content and move the budget to links, or shut it down.

Rule 9 is the hardest and the most important. The shared behaviour across all four sites was adding content and waiting for a miracle, because "write a few more" is always easier than "admit this is not working." Defining a kill condition is not pessimism. It moves the stop-loss from a feeling to a number.

What we are doing with the four

Site A's row is the counter-intuitive one and it is the article's conclusion: a site whose content is right and whose average position is 31 does not need a 79th article. It needs external links. Continuing to write is solving a problem that is already solved.

One more lesson: we could not reproduce our own numbers

Every figure above carries "pulled 9 September 2026". That is not decoration.

We first counted these four sites on 8 September. The same window returned 28,692 impressions and 37 clicks. Re-pulled the next day: 33,615 and 48. The per-site directions do not even agree. Site A went from 22 clicks and 25,927 impressions up to 33 and 31,884, while site B fell from roughly 900 impressions to 390 and site C from roughly 1,700 to 1,177.

We tried six neighbouring windows to reproduce the old figures. Site A lands between 27 and 35 clicks in every one of them; none returns 22. dataState: final and all agree. So this is not a window-definition problem. Search Console restated the data.

Two things follow. The obvious one: always state the pull date on a Search Console figure, because the same query returns different answers on different days. The uncomfortable one: had we run this count a day earlier, this article would carry a different number in its title, and exactly the same conclusion.

So a single pull is not a fact. Keep the script, not the number. Every externally-quoted figure of ours now maps to a script and a JSON file; changing the number means re-running it, and re-running it means updating the date. An article claiming its numbers are checkable, that cannot itself re-run the check, is not making that claim honestly.

This is not an argument that content does not matter

The easy misreading is "so skip content and buy links." That is wrong and more expensive.

The correct reading: content is necessary and not sufficient.

• Without content, links have nothing to point at.
• With content but no authority, the content ranks where nobody looks. That is site A.
• With authority but no content, you get traffic and cannot hold it.

The operating order is: establish what difficulty your domain can actually reach, do the best possible content inside that range, and spend the capacity you saved on solving authority. Site A's mistake was not its content. It was spending 78 articles' worth of capacity on content and none on the constraint that was actually binding.

What to do with this

1. Compute your own number. Open Search Console, pull 90 days, divide clicks by pages. If the result is uncomfortable, the nine rules apply to you.
2. Count your zero-impression pages. Site D was 48 of 54. That ratio tells you directly whether you are reaching anything.
3. Check whether your "verified" claims have links. Five minutes.
4. Define the kill condition now, while you have no emotional investment. It is much harder later.

The tooling side of the same audit is written up in three things your SEO checker cannot see. The vertical-specific version of these constraints is in our complete iGaming SEO guide and the compliance pitfalls piece; how the same money would be spent on ads instead is in the affiliate ROAS benchmarks.

If you would rather hand the whole problem (reach assessment, topic sizing, weekly measurement) to somebody who has already made these four mistakes, that is our SEO site build and management service.

FAQ

How long does SEO actually take?

These four sites suggest the question is wrong. Time is not the variable; the gap between your authority and your target keyword's difficulty is. Site D was 10 weeks old, site C 11 weeks, sites A and B over six months, and the outcome was equally bad for all four. A small gap produces clicks in weeks. A large gap produces nothing in three years.

My site has content but no traffic. What is the most likely cause?

Separate two cases. Impressions but no clicks: you are ranking too deep (past position 20 the rate approaches zero) or the format is being absorbed by summaries. No impressions at all: you are not reaching the SERP. Count your zero-impression pages first — ours was 48 of 54 on one site.

Does great content guarantee rankings?

No. Our best site has five timed withdrawals, four physical handsets and a fifteen-platform test table, and averages position 31.4, while its most-clicked page is a comparison piece with no first-hand data that happens to rank 7.6. On a domain without authority, content quality is not the binding constraint.

Was AI-written content the cause of these failures?

No. Google's position is that how content is produced is not the criterion↗. The actual causes were targeting unreachable keywords, cloning one template four times, and fabricating first-hand claims on one site. Humans do all three unaided.

Do fabricated testing claims actually get caught?

Yes, and by readers before search engines. Search Console showed quoted searches for our invented author's name and for article headlines — that is verification behaviour. Once those queries appear, the exposure is already public.

How long before you can judge a new site?

We use 12 weeks, and we write the kill condition down before launch. The number matters less than defining it while you still have no emotional investment. None of these four had one.

Position 10 and no clicks — should I rewrite the title?

Check the query type first. For a pure definitional query the answer is already on the results page and a better title changes nothing. If the query is operational or comparative and the rate is still low, then it is a title and snippet problem. Also confirm the impressions are human: automated query patterns produce impressions that never convert.

Are you shutting all four down?

No, and the distinction matters: "shut down" and "stop adding content" are different actions. One stops production pending a decision, one fixes two factual errors and waits, one gets version control and more time, and one stops writing entirely and moves its budget to links.

---

Explore our advertising strategy & media buying services →