1. Agents that act, not just answer — prepare now, lightly
The current generation answers questions. The next books, buys and fills in forms. Three competing standards emerged through 2025 and 2026: UCP (Google and Shopify, for discovery and cart), ACP (OpenAI and Stripe, for in-chat checkout) and AP2 (now governed by the FIDO Alliance, for proving who authorised a payment). They are designed to stack rather than compete outright. TIER B Google's own documentation now points site owners at agent-friendly site practices. TIER A
What to do now: if you sell products online, keep your product data clean, structured and accurate — that is the input every one of these protocols consumes, and it is work you should be doing anyway. If you sell services, make sure your enquiry form works without JavaScript gymnastics and that your key information is in plain HTML. Do not implement a protocol yet. Three competing standards means at least one will lose.
2. Someone else deciding whether AI can read you — watch closely
Cloudflare's shift through 2026 from pay-per-crawl toward paying based on whether content was actually used, and its move to block mixed-use crawlers by default on ad-bearing pages from September 2026, is the visible edge of a much larger negotiation between publishers and AI companies. TIER B
What to do now: know what your infrastructure provider's defaults are and check them after every platform change. The risk is not that you make a bad decision — it is that a decision gets made for you by a default setting and you find out six months later. Add "check bot-management defaults" to the quarterly routine.
3. Measurement getting real — prepare now
Search Console's generative AI reports launched in June 2026 with impressions only. Clicks, queries and richer breakdowns are stated as coming. TIER A When they arrive, the guesswork drops sharply and the vendor tools that currently sell inference will have to compete with free first-party data.
What to do now: make sure Search Console is properly verified for every property you own, including subdomains, so you have history when the data widens. Do not sign a long tool contract on the assumption that this gap stays open.
4. The commodity content collapse — act now
This is the shift with the clearest implication. As models get better at synthesising, the value of content that merely explains something already explained falls towards zero. The AI simply answers it. What stays valuable is what a model cannot generate: your operational data, your first-hand experience, your judgement about edge cases, your specific local knowledge.
What to do now: stop investing in explanatory content that competes with the model, and move that budget into the things only you have. This is Chapter 10, and it is the single most future-proof position in this book.
What will not change
Worth ending here, because it is where your effort is safest:
- Machines will need to read you. Whatever the format, being mechanically readable will always be prerequisite.
- Corroboration will always beat assertion. Any system that assembles an answer from multiple sources will reward agreement between them.
- Specificity will always beat vagueness. Numbers, names, conditions and sources are what makes a claim usable.
- Being genuinely good at something, and demonstrably so, will keep working. Unfashionable, and the only strategy that has survived every platform shift so far.
Shift budget out of content that explains things and into content that proves things. That single reallocation is the correct response to all four shifts at once.