This report is late. The 30-day checks on our guest post pilot were due from 27 September and we ran them on 1 October, so each placement was checked between day 30 and day 34 rather than on the day. We said we would publish this whatever it showed, and on time. We managed the first half.
The pilot, briefly: between 28 August and 1 September we bought ten guest post placements for this site from a screened shelf, plus one bought directly from a reseller, and split them into two arms decided in advance. For five, we submitted the host page for indexing ourselves. For the other five we did nothing and left it to the publisher. The protocol was published before anything was ordered.
What the check tested
For every placement: is the page still live, is the link still there, is it still dofollow, is the anchor text unchanged, and is the host page in Google's index. A placement passes only if all five are true. Every result is in the public pilot log, as structured data.
The results
| placements | all five checks pass | |
|---|---|---|
| submitted the host page | 5 | 5 |
| left it alone | 5 | 3 |
| bought direct from a reseller | 1 | 1 |
Every link we could check is still live, still dofollow, and still carries the anchor we asked for. Nobody quietly removed or nofollowed anything in the first month, which is the most common way bought links fail.
The two that did not pass, both in the arm we left alone:
One host page dropped out of Google. It was indexed on day 20. At day 33 it was not found on two separate checks. The link on it is fine. The page it sits on is, for now, not somewhere Google is looking.
One host blocked our checker. The site served a bot verification page instead of the article, so we could not see the link or test the index. That is not evidence the link was removed, and we have recorded it as unverified rather than failed in spirit.
The indexing timeline is the finding we care most about:
| arm | host page indexed after |
|---|---|
| submitted | 3, 6, 6, 7 and 9 hours |
| left alone | 6 hours, 11 days, 20 days (then dropped), about 34 days, unverified |
Submitted pages were indexed within a working day, every time. Pages left alone ranged from six hours to more than a month, and one that made it in did not stay. We wrote up the mechanism in do backlinks need to be indexed?.
Did the links move anything?
Here we have to be careful, because the honest answer is mostly no, or at least not measurably.
| target page | before (28 days) | last 28 days |
|---|---|---|
| homepage | 7 impressions, position 45 | 80 impressions, position 9 |
| index checker | none | 2 impressions, position 71 |
| indexing statistics | none | 3 impressions, position 41 |
| methodology | none | 1 impression, position 3 |
| marketplace study | 4 impressions | none |
| other targets | none | none |
The homepage moved a long way, from page five to the bottom of page one. Four placements point at it. So does almost everything else we did this month: an entity page, brand searches, posts, mentions. We cannot separate the links from the rest, and we will not claim we can.
Every deeper target page got one link each, and one link each did nothing visible in Search Console in a month. That is a result worth publishing: a single guest post on a mid-tier site does not, on its own, move a page that nobody was searching for.
What we got wrong in choosing them
We chose these hosts with a screen that judged a site largely by its homepage. This month we rebuilt that screen to read what sites actually publish: their sitemaps, their recent posts, what Google has indexed from them, and the pages themselves. Run against our own pilot hosts, the new screen would have rejected several of them. The method, and what it does to a shelf of 159,736 listings, is in how the screening works.
Every purchase from here goes through the stricter screen. These ten stay in the experiment, unchanged, because changing the sample halfway would make the 90-day results worthless.
The rest of the month
Five weeks after launch, on a domain with no history, the free tools have been used for 10 SEO audits, 5 competitor reports, 48 index checks, 31 keyword runs and 17 AI visibility checks. Seven people other than us have joined the weekly email. We had our first visitor referred by ChatGPT, who signed up within seconds, and a consultancy that ran four of the tools in one sitting. Search Console recorded 1,594 impressions across 166 queries and five clicks. Small numbers, all real, none bought.
What happens next
We are buying more placements and running the same measurement. The rules for the next batch are already written: every host must pass the stricter screen, the links point at our homepage with brand anchors, and the arms are decided before anything is ordered. The 90-day check on the original ten runs on 26 November and will be published that week, whatever it shows, and this time on the day.
What this proves, and what it does not
It shows that ten bought links, from a screened shelf, all survived their first month intact, and that submitting the host page for indexing got every one of them into Google within hours while leaving them alone took days to weeks and sometimes failed.
It does not show that the links helped us rank. One page moved and we cannot attribute it. The sample is eleven placements on one site in one month; one treatment placement (P01) was indexed before our submission could have acted, so the clean treatment arm is four; one control could not be checked; and our index checks can miss pages, so "not found" is weaker evidence than "found". Everything is in the pilot log at the pilot log., including the parts that went wrong.