The 147 reading row we know least about, and exactly what we know
searchengineinfo.org reached 147 of 1,260 readings without being an operator, a suite or a publisher, and here is the full inventory of what our own data says about it.
Published by AI Knows Us · Updated 19 October 2026
searchengineinfo.org was a source in 147 of the 1,260 readings we took from ChatGPT and Gemini on 16 October 2026, which puts it tenth in that source table beside four Google hostnames, two OpenAI hostnames, a code host, an established search suite and a technology newsroom. It is the row in the whole table that we can say the least about, and that makes it the row most worth setting out honestly rather than explaining away.
What follows is everything our own readings contain about that host, and then the list of things they do not contain. The second list is longer, which is the point.
What we measured
- 147 of 1,260 readings cited the domain, on 16 October 2026, counted as one domain in one reading and deduped.
- It is classified as a vendor or agency own site, the same classification as the search suite four rows above it at 199 readings.
- It was one of the 26 domains whose robots.txt we fetched on 20 October 2026, drawn from the ten most cited hosts of this run, six named specialist competitors, the adjacent suites and directories our data surfaced, four small agencies our audit named as local winners, and our own domain.
- It appeared in our access run and then changed our rules. On one reading it looked like a host refusing AI crawler agents by name. Hours later it returned 200 with the full article to almost every agent, desktop browser included, and returned an explicit HTTP 429 under GPTBot. It was a rate limiter, and the finding was withdrawn.
That withdrawal is worth more to a reader than the finding it replaced. Of the 102 build spec conversations run on 15 October 2026, 68 asked for failures and negative results to be shown, and a retraction with its cause attached is the hardest version of that to publish.
What we do not know, listed individually
| Question | What our data says |
|---|---|
| Why was it cited 147 times | Nothing. The run records which domain was in a source list, not why |
| Which questions pulled it in | Not published at the per question level for this host |
| Which engine cited it | Not broken out for this host in this run |
| How old or how large the site is | Not measured anywhere in our readings |
| Whether it is cited in other categories | Not measured. Our bank is one category, ours |
| Whether a new site could reach the same count | Not measured. Nothing in our data tests that |
A reader who wanted this row to be a success story could fill every one of those cells with a plausible sentence, and none of it would be measurement. The cells stay empty.
The one inference the table does support
There are two rows of the same classification in that top ten, at 199 and at 147 readings, and only one of them is a name a buyer in this market would recognise. Whatever drives these counts, recognition alone is not sufficient to explain the second row. That is a narrow conclusion and it is the only one available from a count.
It also sets the shape of the useful question, which is not how this host did it. The useful question is what your own 147 would be made of, which is answerable only by taking your own reading on your own questions and writing down the date.
What a reader should do with a row like this
- Treat an unfamiliar domain in a source table as an open question, not a finding. Open two of the cited addresses before you conclude anything.
- Establish that the host publishes anything of its own. 712 of the addresses in this same run belonged to our data provider's own fetching setup rather than to any publisher, and they were pointed back at the pages they serve before any share was calculated.
- Record what you do not know. An empty cell in your own table is a research task. A filled cell with a guess in it is worse than the gap.
We are an interested party here: we sell the measurement that produced this table and the publishing work that follows it, which is exactly why the unknown row gets printed with its unknowns rather than turned into a case study. Our own count in the same run was 6 of 1,260. The search console side of our own record, with both windows and the export date on it, is on our proof page.
Three ways a row like this normally gets explained away
Each of these turns up in commentary about citation tables, each sounds reasonable, and none of them is supported by a count.
- It must be an exact match domain, so the name did it. A hostname cannot be tested as a cause by a table that records hostnames. The claim needs a controlled comparison and this run is not one.
- It must be thin pages at volume. Our readings record no page counts, no word counts and no publication dates for any cited domain, so there is nothing here to support or deny that.
- It must be a technical trick. The one technical reading we do have about this host is the withdrawn one, where an apparent per agent refusal turned out to be a rate limiter with a 429 behind it. That is a reason to be slower with this host, not faster.
Each explanation is attractive because it converts an unknown into a tactic. The cost of accepting one is that a reader who checks finds nothing behind it, and then stops trusting the rows that were measured properly.
Questions about an unfamiliar domain in a citation table
How many readings cited searchengineinfo.org?
147 of 1,260, read on 16 October 2026 from ChatGPT and Gemini across 630 questions in three markets.
Do you know why it was cited so often?
No. The run records the domain in the source list and nothing about the cause, and the cells above are left empty rather than filled with a plausible explanation.
Was this host blocking AI crawlers?
It looked that way on one reading and it was not. A later reading returned the full article to almost every agent and produced an HTTP 429, so the apparent block was a rate limiter and the finding was withdrawn.
Is it a big site?
Our readings do not measure the size, age or traffic of any cited domain. What they measure is how many readings had it in the source list.
What does its presence prove about small sites?
Only that the same top ten holds a widely known suite at 199 readings and this host at 147, which rules out recognition as a complete explanation. Everything beyond that is unmeasured. The design and full source mix of the earlier run in this programme are published as 597 readings across ChatGPT and Gemini in three markets.
The row that teaches the most is the one with the gaps written in
A source table is useful in proportion to how clearly it marks its own edges. This row is at the bottom of our top ten, it has the fewest facts attached, and setting out the empty cells is what lets somebody else take it further rather than quote it wrongly. When you build your own table, keep a column for what you have not established. That column is the research plan.