Insights
Insights on AI search,
written while we measure it
Writing about search visibility while we build our own, in public. Every piece here is either a mechanism we can demonstrate or a measurement we are taking on our own domain — with the date it was taken and the method that produced it attached to it.
01 — Published
Everything in this section
Four pieces so far: two explainers on how answer engines pick and time the passages they cite, one buying guide, and the running log of what happens when we run our own service on our own domain, starting from an audited zero. A fifth, written in Spanish rather than translated, lives in the Spanish section.
How Long It Takes to Get Cited in AI Search
Not one interval but three clocks — recrawl, index refresh and corroboration — and only the first is yours. Eighteen days of daily measurement on our own domain, day by day against what shipped, published with the zeros in.
Best Generative Engine Optimization Agencies, and How to Read a List Like This One
Five agencies compared with the selection criteria declared before the entries, what we verified separated from what each firm claims about itself, and why almost every list like this one was written by an agency that appears in it — including ours.
August on August: Measuring Our Own AI Search Visibility
We sell AI search visibility and our own site had none. The audited baseline, every change made since, and fifteen consecutive daily measurements that have not moved — published with the zeros in.
How AI Answer Engines Choose Which Sources to Cite
Answer engines retrieve passages, they do not rank pages. What that changes about how a page has to be written, the four properties the cited passages share, and how to measure progress before the first citation appears.
02 — How we publish
Every figure carries its date and its method
There is one editorial rule and everything here follows from it: no figure appears without the date it was taken and the method that produced it. It sounds like a technicality. It is the whole difference between a report you can act on and a report you can only believe.
A number without a date cannot be compared to the next one, so it cannot show a trend — and a trend is the only thing anyone is actually buying. A number without a method cannot be reproduced, and a number nobody can reproduce is an assertion. Both failures look identical on a slide: a figure, large, in a box. You only find out which one you were given when you try to use it.
The rule has a consequence we did not particularly enjoy. Applied honestly, it means publishing the measurements that did not move, and most of ours have not: sixteen verification rounds between August 11 and August 28, 2026, in which our domain did not appear once. Those sixteen readings are in the log with the same prominence as anything else, because they are the part that carries information. Sixteen identical results say the result set is stable, which rules out the variable we were changing. One flattering reading, published alone, would have ruled out nothing and implied a lot.
It also decides what never gets written. There is no piece here reacting to someone else's product announcement, and none built to occupy a keyword we have nothing first-hand to say about. Not out of restraint — it is that a piece like that has no dated measurement to attach, so under the rule there would be nothing to put in it.
If you are evaluating an agency — us or anyone else — the rule works as a test you can run in a couple of minutes. Take any figure in any deck you have been sent and ask for the date and the method. The answer arrives quickly or it does not arrive.
03 — Questions
Frequently asked
What is published in this section?
Three kinds of piece and nothing else. Explainers, which describe a mechanism we can demonstrate — how answer engines select the passages they cite, for instance — logs, which are the running record of a measurement we are taking on a fixed cadence, and buying guides, which compare suppliers with the selection criteria stated before the entries. There is no news, no commentary on someone else's announcement, and no piece written to occupy a keyword we have nothing to say about. If we have not measured it or done it, it does not go up.
Why do the articles publish results that did not improve?
Because a measurement that only gets published when it moves is not a measurement, it is a selection. Our own log has sixteen rounds between August 11 and August 28, 2026 with the same result on the tracked queries, and those sixteen identical readings are what let us rule out a cause: the result set is stable, so the variable we were changing is not the one that controls it. A single flattering reading could never have told us that.
Where do the numbers in these articles come from?
Every figure carries the date it was taken and the method that produced it, and there are three methods in use. One is live search-result verification: we run the query, read the results and record whether the domain appears. Another is a crawl of beaugust.co with a search-engine user agent. The third is Search Console, whose account was opened on August 26, 2026 and so far reports a two-day window. The three are recorded separately and never averaged into each other.
Is this section in Spanish as well?
Yes, since August 28, 2026, at its own URL — and it does not list the same pieces. The Spanish section carries what was written in Spanish rather than translated, which for now is one buying guide, and it links these four with a label saying they are in English. The reason is a measurement: across twelve queries run between August 19 and 26, 2026, the sources these engines retrieve in Spanish shared no platform at all with the English ones. A translation would send a Spanish reader to a board their game is not played on.
Find out what the engines say about you
The log is updated as measurements come in, so the fastest way to see how this works is to read the one that is still open.