I measured the link value of 101 directories and launchpads. One free dofollow link came out of it.

Safak Tepecik — Tue, 08 Sep 2026 13:13:39 +0000

Everyone who ships a side project eventually reads the same advice: submit it to directories, list it on launchpads, get your backlinks. I wanted to know how much of that advice survives contact with the actual HTML, so I spent a few days measuring instead of guessing. The result is a dataset of 101 platforms, and it is less encouraging than the advice suggests.

What I actually measured

A backlink only passes authority if four things are true at the same time, and most write-ups on this topic check exactly one of them. So I checked all four for every platform.

The first layer is the rel attribute on the outgoing link itself. A link carrying nofollow, sponsored or ugc is a visit, not an endorsement. The second layer is the meta robots element of the page holding that link, because a page marked noindex never enters the index and its links are therefore moot. The third layer is the X-Robots-Tag HTTP header, which can do the same job as the meta element while staying invisible in the page source. The fourth layer only applies when the platform routes outgoing clicks through its own redirector, and there the question is whether that redirect path is blocked in robots.txt.

Two rules kept the data honest. I always measured an existing listing belonging to somebody else, never my own, so that the result describes the platform rather than my account. And when a page could not be retrieved at all, I recorded it as not verifiable rather than inferring a verdict from a screenshot or a blog post.

The headline number

Of the 101 platforms in the dataset, ten carry a genuinely dofollow link on an indexable page, and exactly one of those ten looked reachable without paying or trading something for it. That one came with a condition I did not see until I tested it on my own account, and I come back to it at the end. Thirteen actively devalue the link with a nofollow attribute. Three sit on pages that are noindex outright, so the attribute never gets to matter. Ten deliver no link to the listed vendor's website at all. The detail page mentions the product, links its own social channels and the CDN, and simply never points at the company behind it.

The remaining sixty-five came back as not verifiable, and I want to be precise about what that means, because it is the single most important caveat in this dataset. Seventeen of them answered a plain HTTP request with a 403, which is bot defence rather than an SEO decision. Others are JavaScript applications whose links only exist in the rendered DOM. Not verifiable is not a synonym for bad; it means I refused to write down a number I had not seen.

The pattern behind the paid nine

Nine of the ten dofollow results do not form a group of generous platforms. They form a price list.

StartupBase states in its own comparison table that the backlink is "not included" in the free tier, and the free listing additionally expects a badge on your site. Its paid tier at 39 USD advertises the dofollow link as the product. MicroLaunch sells the same thing at the same price. Fazier wants either 29 USD or a badge plus three public comments on other launches. OpenTools has no free path at all; the submit button is labelled "Continue to Subscription". Onlinestreet, a German directory, gives one link to your homepage for free and three links for 49 euro a month.

So the honest summary of the dofollow column is not "nine platforms give you a link". It is "nine platforms sell you a link, and I measured the version somebody else had already bought." The tenth is the developer profile described further down.

The mistake I made halfway through, and how it was caught

For the first half of the measurements I assumed that a dofollow link found on a stranger's listing proved the platform hands out dofollow links. That assumption is wrong, and I only noticed because of a listing I happened to own.

On Cylex, a German business directory, my own free listing links my site with rel="nofollow". On the same day, two other Cylex listings linked their companies with no rel attribute at all. Same platform, same day, opposite results. The difference is not technical, it is commercial: the free entry gets the devalued link and the paid entry gets the real one.

This means a measurement taken on somebody else's listing tells you what the platform is capable of, not what the free path delivers. Every claim about a free backlink needs a second piece of evidence, namely that the listing you measured was itself free. Half the SEO advice circulating about directories fails exactly here, and mine did too until that one entry contradicted me.

Browser extension stores are not a backlink strategy

This one deserves its own paragraph, because it is repeated constantly and it is measurably false.

The Chrome Web Store links the developer's website with rel="ugc nofollow". The Edge Add-ons store does not put the vendor's website into the delivered HTML at all. Firefox AMO routes the link through prod.outgoing.prod.webservices.mozgcp.net and attaches nofollow on the way out; eleven of the twelve outgoing links on the listing I checked carried it.

Publishing an extension is a fine way to reach users and to distribute your work. It is not a way to acquire authority, and the three largest stores agree on that.

The large knowledge platforms are settled

Wikipedia, Wikimedia Commons and Wikidata devalue every external reference, which is well known but still worth having in a table. Less commonly stated: GitHub, Hugging Face and Zapier do the same. A GitHub repository's homepage field carries nofollow, and so does the link in a Hugging Face model card.

These platforms are still worth being on, because a language model reading the web forms its picture of your project from them. That is a different goal from passing authority, and conflating the two is how people end up disappointed by a strategy that worked exactly as designed.

What did survive

Two free channels held up under the same four-layer test, and both are ordinary rather than clever.

The first is a developer profile on this very platform, and it comes with a catch I only found by making an account of my own.

The website field on an established dev.to profile renders as rel="noopener me", with no meta robots element and no X-Robots-Tag. I checked seven profiles, four of them ordinary community members rather than well-known accounts, and all seven behaved that way.

Then I created a profile, filled in the same field, and measured my own. It renders as rel="noopener me ugc". The ugc is the difference, and ugc devalues the link just as nofollow does. On the same day, re-measuring three of the established profiles still returned rel="noopener me" without it, so this is not a platform-wide change. A new account gets the devalued version; an account with standing does not.

I do not yet know what removes it, and I am not going to guess. The obvious candidates are published posts, account age, or some reputation threshold. That is a measurable question and I will re-measure my own profile once it has posts on it. For now the honest statement is narrower than the one I set out to make: dev.to can give a free dofollow profile link, and it does not give one to a brand-new account.

Hashnode looks similar from the outside and does not hold up. Its profile website field carries nofollow ugc, and four of the six profiles I sampled were noindex at the page level. About.me devalues the link on both profiles I could retrieve.

The second is editorial, and in my case it is also the most relevant one. German university career services publish curated link lists for their students, and those lists do not use nofollow. A careers page at the University of Würzburg links two photographers with rel="noreferrer" only. Hildesheim links about thirty external career resources the same way. Frankfurt's writing-resources page links dozens of third-party tools with no rel attribute at all. None of those pages carries a robots restriction on any of the four layers.

Getting onto a list like that requires being genuinely useful to the students reading it, which is slower than paying 39 USD and considerably more durable.

The dataset

The full table is published under CC BY 4.0 as CSV and JSON, one row per platform, each with the exact evidence URL I measured, the four layers separately, and the verdict. If a row says not verifiable, the reason is in the row. I would rather hand over a dataset with sixty-five honest gaps than one with sixty-five confident guesses.

The measurements were taken on 31 August 2026. Platforms change their link handling without announcing it, so treat every row as a dated observation rather than a permanent property. If you re-measure one and get a different answer, that is a finding, not a contradiction.

I collected this while working on profilfoto-ki.de. The files are at
/downloads/linkwert-verzeichnisse-2026-08.csv
and /downloads/linkwert-verzeichnisse-2026-08.json.

DEV Community: Safak Tepecik