Search

You have something worth finding. Today is how anyone finds it — and this is the channel that has changed most in the last two years, because a growing share of people never see a list of links at all. You have been one of them for twelve days.

1. Search now asks two questions, not one.

For twenty years search meant one thing: does your page appear in a list of links, and how high.

There are now two questions, and they have different answers:

	The question	What wins it
Can people find it?	Do you appear when someone searches	Relevance, authority, the page being the best answer
Does the machine name you?	Does an assistant mention you when asked	Being quotable, being cited elsewhere, having something specific to say

You will meet three letter-sets for this — SEO, AEO, GEO. Being found, being the answer shown without a click, being cited by an assistant. Do not treat them as three disciplines to learn. They are one job with three outcomes, and the fundamentals that serve one mostly serve the others.

2. Intent decides everything else.

The most common search mistake is choosing a query by how many people search it, and ignoring why.

Someone typing "what is social media scheduling" wants to understand something. Someone typing "buffer alternative for agencies" is comparing options with a credit card nearby. Same topic. Completely different reader, and a page built for one fails the other.

The practical rule: match the page type to the intent. An article ranking for a transactional query converts badly, and a product page ranking for an informational one bounces. Both look like a ranking win and neither produces a customer.

3. The click is disappearing.

This is the part that has genuinely changed, and the numbers are worth knowing.

What is not in dispute is the direction, and it is sharp: a large and rising share of searches end without anyone clicking anything, and assistants now answer a great many questions that used to start with a search box. You have done exactly that for two weeks.

Two things follow.

Ranking and traffic have come apart. You can rank first, be summarised above yourself, and get fewer visits than last year. Position is no longer a proxy for people arriving.

For many informational questions, being the answer now matters as much as being a result. For transactional ones — someone ready to sign up — the click still matters and still converts. Match your expectation to the intent.

Either way it makes yesterday's structure — discrete blocks answering discrete questions — a search decision rather than a formatting one.

4. Choosing queries you could actually win.

Three things decide whether a query is worth your time, and most people weigh only the first.

Demand	How many people search it. Necessary, and the least interesting of the three.
Difficulty	Who already ranks. If the first page is five established competitors with a decade of authority, a new site is not winning it this year.
Value to you	What happens if you win. A thousand visits from people who will never buy is worth less than twenty from people with the problem.

For a product with no search authority yet, the answer is almost always specific over large. "Social media tool" is unwinnable and vague. "Schedule posts for multiple clients without paying per seat" has a fraction of the volume, a fraction of the competition, and describes a person who would actually buy.

This is where keyword data earns its place — and only here. It tells you how many people share a question and what words they use. It does not tell you which question matters. That came from Day 7, from real people, and the order matters: real question first, demand check second.

5. The page has to be the answer, not about the answer.

Once you have the query, the job is straightforward and most pages fail it.

Answer the question directly, early, in plain words. A page that circles the subject for four paragraphs before answering loses the reader and gives an assistant nothing clean to quote.

The fundamentals that still matter, briefly:

Notice how much of that is just writing clearly for a person. That is not a coincidence, and it is why chasing tricks ages badly while clarity does not.

6. Being cited is a different game from ranking.

Here is the finding that surprised me most this week, and it reframes the whole channel.

The consistent finding across the analyses I can find — and the same caveat applies, I have not verified the underlying data — is that the large majority of AI citations come from earned media. Third-party articles, community discussions, places other people wrote about you. Paid and advertorial barely registers.

Read that against Friday's map. The strongest lever on whether an assistant names you looks like the earned column, not the owned one. Being talked about elsewhere may matter more than anything you publish yourself.

Two more things worth holding:

Citations are rare at all. Most assistant answers name no source whatsoever.

The competitive set is different. The domains cited in AI answers for a topic are frequently not the ones ranking in Google for it. So you can be losing in one and present in the other, and you will not know unless you check both.

What earns a citation looks a lot like yesterday's lesson: original data, real benchmarks, named results, a clear point of view. Thin derivative pages do not get quoted, because there is nothing in them to quote.

And none of this replaces the fundamentals. A page still has to be crawlable, indexed, fast, useful and trusted. AI visibility is another way to be found, not a substitute for being findable.

7. Measuring AI visibility without fooling yourself.

This is where most audits wave their hands. Do not.

Ask what a buyer would type, not your own name. "What is LotsSocial" tells you almost nothing — it is navigational and the answer is trivially yes. "Best social media scheduler for a small business" tells you whether you exist in the conversation where a decision gets made.

One reading is an anecdote. The same prompt asked twice can return different answers. Never say "ChatGPT recommends us" from a single call. Say how many prompts you tested and how often you appeared.

Record the conditions or the number means nothing. "Mentioned 2 out of 3 times" is not a measurement unless you also captured what was asked, where, when, and what came back.

And report absence as clearly as presence. Not being mentioned is the finding that matters — along with who was mentioned instead. A baseline of zero is a real result, and it is the one we have today.

8. Ranker — and the line with Scout.

Ranker owns exactly two questions: can people find this in search, and do AI answer engines name it.

Its first instruction is the one that separates an audit from a guess: start with our own measured data. Search Console first — the queries we already appear for, the pages, the impressions, the actual positions. If that is not connected, it is told to say so before anything else, because everything after it is weaker.

Then licensed data for what we cannot see ourselves. Search Console only shows queries you already appear for — it is silent on everything you are invisible for, which is most of them. Demand and difficulty have to come from somewhere else, and from real data rather than estimation.

The boundary with Scout is worth learning as a way of thinking:

When a question is really about positioning or what customers say, Ranker is told to hand it back rather than produce a thinner version from keyword data — because keyword volume is not a substitute for talking to people.

That is a good rule for you too. Search data tells you how many people share a question. It never tells you why they are asking.

9. What this did not cover, and why none of it is fast.

The honest timeline first. A new site publishing good work sees very little for a quarter. Search rewards accumulated trust, and there is no version of this that produces results in three weeks. Anyone selling you one is selling you something else.

That is not a reason to skip it. It is a reason to start before you need it, and to judge it on the right rung of Day 18's ladder rather than abandoning it in week two.

Not covered:

Where to go. Google's own documentation is genuinely good and free. Beyond that: the fastest way to learn this is to watch one page you made, weekly, for three months — which is what next week's measurement lesson sets you up to do.

Tomorrow — Day 14: Social. Distribution and familiarity, not publication — and the uncomfortable part of marketing a social media tool, which is that our own feed is a product demo whether we mean it to be or not.