What I believe about AI and the way we work
A record of the things I think are important. Each one says what would change my mind.
Discovery in AI answers
Nobody can control what ChatGPT says about a company
Anyone who says they can is overselling. The models sample everything they can find and repeat what they find most often and most of what they find is on pages the company does not own. What a company can do is correct the facts at the sources the models cite and publish plain structured information the models learn from and then measure on a schedule whether the answer moved. That is a plan and not an edit. What would change my mind is a brand that moved an answer with a change to its own site alone and showed the before and after.
Know what AI thinks and not just what it says
Counting names tells you whether a brand is in the answer. It does not tell you why. A model has learned some things and finds others when it searches and repeats a third set because they are everywhere. The three need different fixes. A fact the model has learned wrong needs the sources corrected and time. A fact it finds wrong needs the page it finds fixed today. A fact it repeats needs the repetition broken. A tool that only counts names cannot tell them apart and a plan built on it will fix the wrong thing.
Honest numbers
Every figure should show its count and its range. Say how sure you are and what you could not check. Never claim a cause you have not shown. One reading is an observation and a finding is a finding only when it has been repeated. This is the rule the whole site is written to and it is the one I would most like to see the industry adopt.
A model does not read your best page once
It samples everything it can find about a brand and repeats what it finds most often and most consistently. Content for AI is therefore a strategy and not a tactic. One good page is outweighed by the pages that disagree with it and most of those are on sites the brand does not own. The work is to say the same plain facts everywhere the brand already publishes and to be present at every step of the journey and to fix the earned pages too. What would change my mind is a brand that moved an answer with one hero page and showed the before and after.
The way we work
The way we work has changed and most businesses have not caught up
Resource is expensive and AI can now do most of the work with one person leading it. The research on this site is done that way. Teams are smaller and faster and the accountability is clearer because one person has seen the whole piece of work. Most businesses still plan as if a project needs a team of eight and a quarter. What would change my mind is a piece of work of this kind that could not have been done by one person with good tools.
Count what a team still uses a month later
Most AI programmes count what was launched and that number is always good. A routine that is dropped after the demo has cost the team time and given nothing back. The only measure of an AI programme I trust is the share of working days each routine was still used at day 30 and the reason for every one that was dropped. The routine that goes is usually the one with the best demo because it solved a problem the team did not have.
Build one semantic layer and every AI agrees with it
Integrating a business's data is the first step and giving it a structure that means something to the business is the second. The layer that holds those definitions is now a feature of the platforms most businesses already own and nobody needs to buy it again. The work is deciding what each measure means and writing it down once so that the dashboard and the chatbot and the assistant in the spreadsheet give the same number. The measures nobody has defined yet are the ones about what AI and search and people say about a brand and those are the ones I spend my time on.
My details are on the contact page if you want to argue with any of these.