RabbitCodex

Search is turning into one answer, and almost nobody is optimising for it

Ranking tenth on a page of links used to be worth something. When an assistant answers the question instead, there is one source and there is no page. I think this is the most interesting thing happening in my field, and I appear to be almost alone in that.

The short version

Search is collapsing from a list into an answer. When someone asks an assistant a question, one or two sources get cited and the rest are invisible. Optimising to be cited is a different discipline from optimising to rank, and most companies are still doing the second one.

  • An answer cites one or two sources. There is no sixth place.
  • Length was a ranking asset and is a citation liability.
  • Put the liftable answer first, before context, before setup.
  • Half-arrived shifts are the only ones you can be early on without being wrong.

The shape of the change

A results page is a list. Ten links, and being sixth is worth real money, because a person can see all ten and pick the one that looks right. An answer is not a list. It is a paragraph with one or two citations underneath it, and being sixth is worth precisely nothing, because sixth does not appear.

That is the whole thing. Everything else in this post falls out of that one sentence.

Why this is the fun part

Ranking was, for about twenty years, mostly a competition of authority. Links, domain age, the accumulated weight of being old and referenced by other old things. You could not really shortcut it. You could outspend it or you could outlast it.

Citation is a different game, and this is the bit I will corner you about at a party. An assistant picks a source because that source stated the thing cleanly, in a form that could be lifted, with enough specificity to be worth quoting. That is not authority. That is craft. A two-year-old site that answers the question in one unambiguous paragraph can beat a twenty-year-old site that buries the same fact halfway down a listicle, and it can do it this month rather than in 2031.

I have been doing this long enough to be deeply suspicious of anything that sounds like a shortcut, so let me be clear that this is not one. It is a genuinely different discipline that happens to be wide open, because almost nobody is doing it on purpose yet.

What it actually looks like

Concretely. These are the four things I check on our own pages, in this order:

  • Answer first. The question gets answered in the opening paragraph, in one sentence a machine can lift without dragging three paragraphs of setup along with it. If your best sentence lives in your conclusion, it does not exist.
  • One fact per sentence. Compound sentences are hard to quote safely, and a model will skip a sentence it cannot truncate without lying. This one change does more than everything else combined.
  • Numbers and dates, stated plainly. "Recently" cannot be cited. "On 3 September" can be. This is free, it takes four seconds, and hardly anyone does it.
  • Structure that survives being stripped. Headings that are real questions. Lists that still make sense without the paragraph above them. If your meaning lives in your layout, it dies in the pipe.

None of that is exotic. It is much closer to writing good documentation than to anything that was ever called SEO, which is why I suspect engineers are going to turn out to be unreasonably good at it.

The uncomfortable bit, which I will not skip

If you win the citation, the reader may never visit you at all. You got quoted, they got their answer, nobody clicked. Every analytics dashboard on earth will score that as a loss, and it will keep scoring it as a loss for a while yet.

I do not have a tidy answer to that and I am not going to pretend otherwise. What I will say is that the alternative, being neither cited nor visited, is not an improvement, and that the people building measurement for this are running roughly eighteen months behind the people building the assistants. That gap is miserable if you have to report on it every month. It is an opportunity if you can stand to be early.

We are early on purpose. Ask me in a year how that went, and I will tell you honestly, because that is more useful to both of us than a victory lap.

Asked and answered

Is this the same as SEO?
It overlaps, and it is not the same. Being crawlable, fast and correct still matters enormously. Keyword tactics and length for its own sake now actively hurt you, which I enjoy more than I should.
Does it have a name?
Several, none settled. Answer engine optimisation and generative engine optimisation are the two you will see. The naming will sort itself out later than the practice does.
How would I know if it is working?
Ask the assistants questions you should be the answer to, and see who gets cited. That is a cruder measure than a rank tracker and it is the one that matters.
Do I have to choose between the two?
No, and anyone telling you to is selling something. Most of the work overlaps. The part that does not is the part about how you write.

You can be told first

No newsletter. When there is something to see, the people on this list see it before the log does.

Friday 3 min

Friday: what we are not saying yet

The weekly update. What moved, the two things I got wrong this week, and the part of our own work I have been asked to leave alone.

Log 3 min

Announced is not shipped, and the difference costs weeks

Most coverage treats a blog post and a working product as the same event. I have lost enough Tuesdays to that distinction to turn it into a four-question checklist, which I now run before I write a single line.