Chris Green: Patents, Bots & the Cannibalization Reversal

Fifteen years of SEO theory was built on Google’s patent literature. Chris Green of Torque Partnership takes the two load-bearing ones, PageRank and Reasonable Surfer, and asks what survives a web where most of the traffic hitting your pages was never a person. He is also the sceptic on markdown companion files, and the sceptic is the useful voice in that argument.

Key takeaways

• Patents are principles to reason from, not rules to obey. Simulating link equity is a fine way to picture a change. Betting a rollout on the simulation is the risk.
• Agent traffic is polluting the analytics your budget rests on. Worst when the agent rides a real user’s browser, session and device, where the usual bot tells are absent.
• Markdown companions are near-free on a small site and a governance problem on a big one. The cost is not authoring. It is who owns the drift.
• Nothing on the internet will tell you your markdown is wrong. Search Console flags broken HTML. No equivalent exists for a companion file.
• An isolated client test found three to four selectable links to be a citation sweet spot. He labels it as isolated and does not oversell it.

On this page

• Patents as principles, not blueprints
• What agent traffic does to your data
• The sceptic’s case on markdown companions
• The cannibalisation reversal, and access logs
• Chapters and timestamps
• People, ideas and sources mentioned
• Questions this episode answers
• Go deeper

Patents as principles, not blueprints

We know when the patent comes in, we know when it’s established. We don’t often have a consistent concrete understanding of for how long that mechanism is at work or what other systems intertwine with it... learn them, be aware of them, be conscious of them.

— Chris Green

The distinction he is drawing is between using a patent to picture a system and using one to justify a decision. You can move PageRank around a site convincingly in a simulation and change visibility by nothing at all.

If you know how to test it and you know how to be skeptical then it’s still valuable, still highly valuable.

— Chris Green

What agent traffic does to your data

User behavior collapses, like things like dwell time, click frustration points, call to actions, everything like that is turned on its head... it pollutes all of my other data, which then ruins decision making, which then eventually potentially ruins my budget and my remit.

— Chris Green

This is the part of the agentic shift that arrives before any of the interesting parts. Agents do not pogo-stick, do not hesitate, and do not read. Every behavioural metric you use to argue for investment quietly stops describing humans, and nothing in the dashboard tells you when that happened.

The hardest case is not a declared crawler. It is an agent operating inside a real person’s browser, on their session and their device, where the ordinary detection signals simply are not present.

The sceptic’s case on markdown companions

A relatively small static site... like a WordPress or something that’s fairly simple... it’s almost a non task, you can kind of do it automatically, never think about it again. And I think if that’s you and you can do it now, I can think of no reason why you wouldn’t... for a 50,000, 100,000, million page website that’s made with duct tape and sticky tape, it’s just not going to fly.

— Chris Green

Note that this is not a no. It is a scale-dependent yes, which is the answer most of the argument has been missing. His actual worry is drift: what happens when the markdown and the HTML disagree, and which team is accountable when they do.

We end up at a route where a markdown file is serving different data. No one’s gonna tell you at the moment, there isn’t a single bot or service out there that will say, hey, your markdown files are wrong.

— Chris Green

That is the missing feedback loop, and it is a real one. Every other publishing surface you own has something watching it.

The cannibalisation reversal, and access logs

If you had three to four links that were selectable, you were much more likely to get included in the citation than if you had under three or interestingly over that four or five mark.

— Chris Green

One relatively isolated client test, presented as such. The operational consequence is heresy by the old rules: keep the subdomain page, add the main-domain page, and link them, rather than consolidating one into the other.

There is a possibility that we as an industry may have got a little oversensitive to cannibalization in a very kind of strict kind of way. There’s always the potential that it maybe wasn’t the biggest issue.

— Chris Green

The evidence nobody is bringing

One of the other treasure troves of information is access logs. So if we can say these AI bots are visiting on these days, here’s your average count for the amount of times that a user driven bot has appeared versus a training bot versus a search bot... very few people are bringing that kind of value.

— Chris Green

The split that matters is user-triggered against training crawler, because the first is a proxy for a real person being shown your page. He also warns to check the IP range before you trust the user agent, since a spoofed ChatGPT hit will happily contaminate the whole analysis.

Chapters

People, ideas and sources mentioned

Questions this episode answers

• Do Google's old patents still apply to how search works today?
• Does the reasonable surfer patent apply to bots and agents?
• Should I publish markdown versions of my pages?
• How is agent traffic distorting my analytics?
• How many links should a page have to maximise its chance of being cited?
• Was the SEO industry too strict about keyword cannibalisation?
• How do I tell a real ChatGPT crawler from a spoofed one?

Go deeper

Every link above goes somewhere different. These are the ones not already mentioned above.

• the case for the companion file, argued from the opposite side, where Dan Kurtz treats bots as an audience with a format preference and calls the downside bounded.
• a third position, serving markdown by user agent at one URL, which sidesteps the duplicate-URL objection and introduces a cloaking one instead.
• the speaker-labelled text edition of this conversation, written so a model can read it without guessing who said what.
• a step-by-step routine for reading AI bot hits in a raw log, if the access-log section left you wanting the actual procedure.

Watch the interview

Find Chris Green

chris-green.net, his working notes on Substack, and LinkedIn.

Cite this episode

Green, Chris. Interviewed by Jeremy Rivera. “Do Google’s Patents Still Hold Up in an Agentic Web?” The Unscripted SEO Podcast, 6 August 2026. https://unscriptedseo.com/chris-green-do-googles-patents-still-hold-up-in-an-agentic-web/