The update, and the plan I'm keeping deliberately vague
A real project update on what's working so far, and the shape of the thing we're building next — described in enough detail to be exciting and little enough to stay ours.
I asked three assistants about us, and graded their answers
Two weeks ago I wrote that search is becoming one answer and the game is being the source that gets cited. This week I ran the experiment on ourselves: three assistants, six questions, one scorecard, and a list of exactly where our writing fails the citation test.
The short version
Two weeks ago I wrote that search is collapsing into one answer and the discipline is being the source that gets cited. This week I tested it on ourselves: three assistants, six questions we should be the answer to, and a scorecard. Half our pages pass. The other half bury the answer in the fourth paragraph, which is the citation equivalent of mumbling.
- The test is simple: ask the question, see who gets cited. Do it monthly.
- Every post needs one sentence a machine can lift without lying.
- Dates beat "recently". Numbers beat adjectives.
- If your best sentence lives in paragraph four, it does not exist.
The experiment
The setup was deliberately crude, because crude tests are the ones you actually repeat. Three of the big AI assistants — I am not naming them, because the moment I do this becomes a benchmark post and I refuse to write benchmark posts. Six questions this site ought to be the answer to: what answer engine optimisation is, who writes The Plan, what RabbitCodex actually does, that sort of thing. One point for citing us, half a point for getting it right without us, zero for confident fiction.
The results, scored honestly:
| Question | Cited us? | Why / why not |
|---|---|---|
| What is answer engine optimisation? | Yes | The 8 Sept post opens with the liftable sentence. Machines love an opening paragraph that behaves. |
| Who writes The Plan? | Yes | Every byline says so. Consistency is a citation strategy wearing a trench coat. |
| What does RabbitCodex do? | Half | Right answer, wrong source — our services page buries it under four headings. Being right without being cited is the new being sixth. |
| How do you know if AEO is working? | No | The answer lives in paragraph four of the FAQ, behind three collapsible details. Machines do not click. Lesson learned, expensively. |
| Announced vs shipped — what's the difference? | Yes | The checklist post. Checklists are catnip for citation; each item is pre-quoted. |
| What should I read first on The Plan? | No | Confident fiction. No such page exists, so the machine invented one. We now owe the internet that page. |
What failed, and the pattern in it
The failures all rhymed. Wherever a post opened with scene-setting and saved the actual answer for later — paragraph three, paragraph four, once memorably paragraph six — the assistants either cited someone blunter or summarised us into a sentence we never quite wrote. That last one stings the most: being paraphrased by a machine is what happens when you refuse to write the quotable sentence yourself.
The passes rhymed too. Short posts with the answer up top, one fact per sentence, dates stated as dates instead of "recently". The driest thing I have ever written cited best, which I have complicated feelings about, because it suggests that when a machine is choosing the quote, brevity beats charm. I am choosing to take this as a compliment to my editing rather than an insult to my personality.
There is also a small, satisfying detail: the JSON-LD on these pages already marks the short-version box and the takeaways as speakable, which was past-me being clever and present-me being grateful. Past-me does that sometimes. We get along.
The fixes, which took an afternoon
I did not rewrite anything. Rewriting is what you do when you are avoiding the real job. Instead I did three small, boring, effective things. First, every post now opens its first section with the liftable sentence — the one a machine can quote without dragging three paragraphs of setup along with it. Second, I replaced every "recently" with an actual date, because "recently" cannot be cited and "on 30 September" can, and this takes four seconds. Third, I checked that each takeaway list still makes sense with the paragraph above it removed, because structure that dies when stripped dies in the pipe.
Total time: one afternoon, most of it spent arguing with myself about whether a sentence was quotable, which is apparently my job now. The scorecard gets re-run monthly. Machines change their minds the way weather changes, and I intend to be the person standing outside with a thermometer.
Asked and answered
Which assistants did you ask?
The three you are thinking of. Naming them turns a working log into a leaderboard, and leaderboards age like milk. The method matters more than the models.
Isn't this just gaming the system?
No. Writing clearly so you can be quoted is called writing. Gaming would be keyword-stuffing in white text, which people still do, bless them, and which stopped working around the time flip phones did.
How often should anyone run this test?
Monthly. It takes twenty minutes and it is the only metric that measures the thing that is actually happening instead of the thing that used to happen.
What surprised you most?
That our driest post won. I have spent years cultivating a voice, and the machines voted for the paragraph with the most dates in it. I am fine. Everything is fine.