After three articles about what doesn't work, the other side is owed: what does.
The answer is less exciting than anyone would like, which is exactly why it is worth more.
The uncomfortable finding
A research team built a controlled benchmark to compare two things: the new techniques, designed specifically to get a document into conversational answers, against the old work of making a page better so it ranks higher.
The result favoured the old work. The study concludes that most of the new methods are not only largely ineffective but frequently have a negative impact on document ranking, while traditional strategies, the ones aiming to improve the ranking of the source, are significantly more effective.
It makes sense once you think about the mechanics. Before writing, the assistant searches and pulls a few pages. If your page isn't among the ones it pulls, no writing technique saves it. And what decides which ones it pulls looks a lot like what used to decide the order in a search engine.
And the finding nobody puts in a proposal
The same study measured one more thing, and this is what changes the budget conversation.
They simulated what happens when it isn't one company doing this but several. As the number of adopters grows, the gain for each one shrinks. The authors describe it as a congested, zero-sum problem.
Put simply: it is a race where, if everybody runs, nobody gains ground. That doesn't mean you shouldn't run. It means anyone promising you a permanent advantage from this is promising something the field itself says dilutes.
So where should the money go?
Our reading, ordered from most evidence to least.
First, make the page readable. It is the only item on this list that is a yes or a no, verifiable in minutes, and dependent on nobody else. No AI crawler runs JavaScript, so a site assembled in the browser isn't in the conversation at all. This doesn't compete with anything: it is the door.
Second, the old work, done seriously. Pages that answer a real question, with the topic in the title and the answer in the first paragraph. Content kept current. It is what the benchmark found most effective and, incidentally, what still brings 42.75% of web traffic, against the 0.32% AI platforms bring today.
Third, get named elsewhere. An assistant builds its answer from several sources, and almost none of them is your own site. Serious directories, real reviews, press, mentions. It is the slowest and most expensive part, and it is what truly separates a company that shows up from one that doesn't.
Fourth, measurement. At the end, not the beginning. It tells you whether the above worked. It doesn't tell you what to do.
The part nobody wants to hear
If a company hasn't sorted out the basics, this is not where to start.
A site that doesn't explain what it does, a Google listing with no photos or hours, zero reviews, no mentions anywhere off-site: from that starting point, buying AI monitoring is paying to watch a zero every month.
There is a natural order worth respecting, because it makes everything else cheaper: first make it readable, then make it say something, then get others to say it, and only then measure.
We walked that road on our own site before selling it to anyone. What that looks like from the outside is in Problems we solve, and how we work on it is in Get recommended by AI.
