Five Expensive Mistakes

Every one of these was defensible on the day it was made. That is the point of the chapter: nobody commits an error that looked stupid at the time, so a list of obviously stupid mistakes would teach you nothing.

Ranked by what they cost.

1. Starting link acquisition in month seven instead of month two

The largest strategic error in the entire project, and the one with the longest tail.

Seven months went into the lever that was working, building and optimising pages, and nothing into the lever that would determine the next phase. Because links have the same one-to-four month settlement delay as everything else (Chapter 14), the referring domains being pursued now should have started compounding in February.

The ceiling was predictable. On-page work has an obvious end state: eventually every page is as complete as you can make it. That endpoint was foreseeable from month two, and the off-page work could have been running in parallel the entire time at a cost of thirty minutes a day.

The general lesson: start the slow-compounding channel before you need it, precisely because it is slow. By the time you need it, you have added its lead time to your timeline.

2. Ninety blog posts

Ninety-four indexed blog pages produced 1,913 clicks, 1.71% of all traffic, from months of writing.

Two failure modes, both structural:

They competed with the product pages and lost. Someone searching "how to merge PDF files" is served a page that answers the question and does the job. A blog post about the same query is strictly worse, and having both splits your own signals across two URLs while the search engine picks. In one case a blog post is still cannibalising a tool page on a query worth thousands of impressions a week.

The traffic was the wrong traffic. Someone reading about a topic is in a different mental state from someone trying to complete a task. Task intent converts; topic intent browses.

What ten posts should have been, rather than ninety:

The test before writing anything: could a product page rank for this query instead? If yes, build the product page.

3. The fabricated ratings

For several weeks, four pages carried structured data claiming an aggregate rating of 4.9 stars from 8,500 reviews. There were no reviews. The number was invented.

Why it happened, stated plainly because the reasoning is common: rating stars render as gold stars in search results and measurably lift click-through. Competitors display them. Sitting at position 8 watching tens of thousands of impressions produce almost no clicks, it is very easy to reason that the product genuinely is good, that everyone does this, and that the number is directionally true.

None of that is a justification.

Tolkien spent a thousand pages on this exact temptation and it is worth borrowing the shape of it. The Ring is not attractive because it is evil. It is attractive because it works, immediately, visibly, and in the direction you already wanted to go. Every character who reaches for it is reaching for a legitimate goal by a route that removes the years. And the specific cost is never the one they were bracing for: it takes the thing they were trying to protect.

Gold stars work. That is the whole problem with them. The fabricated number would have lifted click-through, in the direction the project already needed, immediately, and the price was not the stars.

Why it was removed. The practical reason is that fabricated rating markup explicitly violates structured data guidelines and is a well-known trigger for a manual penalty, which does not cost you the stars, it can cost you rich results across the entire site or the site itself. Seven months of compounding, wagered against a small CTR improvement. Terrible odds.

The real reason is worse. The product's entire differentiator is a trust claim, that files are never transmitted, which a user largely has to take on faith. That claim was being made on a page that was lying about something the user could have checked. If the review count is invented, why would the privacy claim be true?

You cannot build a trust-based position on a page with a fake number on it. The two cannot coexist and the position is worth more.

All fabricated ratings were removed. The one genuine rating, 4.8 from 124 real responses, was kept, because it is real.

The general rule: when you have no budget, credibility is the only asset that compounds faster than your content. Links come from people who believe you, and the communities most worth reaching are precisely the ones that verify. This extends to your headline numbers: define them precisely and state the definition. In this book, "250K users" means 249,472 unique users measured by Microsoft Clarity over 25 Dec 2025 – 17 Sep 2026. It is rounded up by 528 people, which at the rate the site grew between August and September is under four hours of new users, and it is stated here rather than in a footnote because a rounded headline you have to look up is exactly the thing this chapter is about. A number you can defend is worth more than a bigger one you can't.

4. A hundred bookmark-network submissions

Covered in full in Chapter 25, and listed here because it belongs in the ledger.

Roughly a hundred auto-approve social bookmarking sites, submitted to in a single run. The tactic has been obsolete since the early 2010s; a quarter of the network turned out to be dead domains; and it is a link scheme by the plain definition, sites with no readers that publish anything submitted to them.

What makes it a mistake rather than merely a waste is that the evidence against it was already sitting in this project's own data. Eighty-five directory links had already moved Domain Rating from 0 to 21 and moved rankings not at all. A worse version of a play that had already been shown not to work was never going to work better.

Cost so far: none that can be measured. No manual action, no datable drop. That is not vindication. It is the ordinary outcome, because Google mostly just ignores these links. The real cost is the tail risk and the fact that it sits beside the fabricated ratings in a profile that is supposed to be selling trust.

The response is in Chapter 25 and it is deliberately undramatic: check Manual Actions, prepare a disavow file, do not submit it without cause, and put the next weekend into the dozen real directories instead.

5. Geography, and a domain chosen in four minutes

The domain uses a country-code extension belonging to a country of roughly 500,000 people, chosen because the .com was taken and it looked clean.

What the data actually shows, and I want to be careful here, because this is a case where the obvious story is not clearly supported:

MarketImpressionsCTRPosition
India951,0695.71%8.12
Malaysia123,2093.33%8.48
United States541,1651.82%11.14
United Kingdom168,2791.75%8.86
Australia89,7901.42%9.34
New Zealand19,1791.20%14.14
The country the domain belongs to2302.17%9.08

The pattern is real, and it has narrowed: developed English-speaking markets now rank one to six positions worse than developing ones, where in the first edition the gap ran to ten. The United States alone represents 541,165 impressions at position 11, the largest single pocket of unrealised demand in the dataset.

But the honest reading is that this data does not distinguish between two explanations. One is that the country-code domain causes geographic mis-targeting. The other is simply that US, UK and Australian results for these queries are far more competitive, because that is where the well-funded incumbents concentrate. The second explanation alone accounts for the pattern, and the fact that the domain's own country produced 230 impressions out of 3.76 million suggests the engine is not aggressively targeting traffic there.

I am flagging this rather than asserting it, because the difference matters: one implies migrating domains, the other implies the ordinary authority problem of Chapter 23, solved the ordinary way.

What is not in doubt is that the domain choice was made in four minutes without checking, and that a competitor with a near-identical name on their market's own country extension consistently outranks this project on its own brand name, with an equivalent link profile.

Rules for choosing a domain, which is infrastructure and not aesthetics:

1. Prefer .com. Pay the premium; it is smaller than the cost of not having it.

2. If not .com, use only extensions treated as generic through common use. Verify before buying.

3. Never use a country extension for a country you don't sell to.

4. Search your intended brand name first. If someone already owns that identity, you are signing up to compete for your own name indefinitely.

5. Consider whether you can take payment on it credibly. Trust is not evenly distributed across extensions.

What was kept, unchanged

About this book

Five Expensive Mistakes is chapter 29 of How to Grow Your SaaS to 250K Users Without Spending on Ads, a playbook on taking one product from zero to 250K+ users on search alone, with nothing spent on advertising. Every claim in it is checked against the real data, including the three findings that contradicted the author.

Continue reading

Free tools that implement this book