I should admit something that will sound strange for someone building in this space. I have never used social media. No LinkedIn, no X, no Insta, no account anywhere - not as a hold-out position, just that it never became necessary in my life. Only recently has our team begun publishing under CLEO’s name, and more recently still under mine. I have opened an X account recently, at the precise moment the work made it unavoidable, and looking at these platforms for the first time with no native habits to unlearn. What I see, I think, is clearer for never having been inside it.
Here is what I saw. I had built the infrastructure side well - sites the engines could read, entities they could recognise. And it was not enough to move the needle for a client. I would get a brand cleanly cited and still watch the engine decline to recommend it. For a while I assumed I had the infrastructure wrong. I did not. I had mistaken one outcome for another.
Ask an AI engine for the best tool in some category and read the answer slowly. Two different things are happening in that paragraph. Sometimes a brand is recommended - named as the answer, with a reason. Sometimes a brand is cited - quoted as a source, with a link beneath. They look alike on the page. They are won by completely different work, and confusing them is the single most expensive mistake in this field.
A citation is a function of infrastructure. The engine reaches for a source to support a claim, and your page is clean enough to lift off the shelf - structured, unambiguous, easy to quote. That is the work I had been doing. That is GEO.
A recommendation is a function of reputation. The engine reaches into the pattern of how a category is discussed - what real people have said, across the platforms it has learned to weight - and concludes your brand is the one to name. It is not quoting a source. It is summarising a consensus it inferred from many voices. And that consensus lives in exactly the places I had never been: Reddit, YouTube, LinkedIn, the forums where practitioners actually talk.
That was the wall. Citation is infrastructure; recommendation is reputation. I had built the first and assumed the second would follow. It does not. A brand can be perfectly citable and never recommended, because the engine can read its page but finds nothing in the open web that treats it as an answer. The reputation has to be earned off your own pages - which means the infrastructure work, however good, is necessary and not sufficient.
I find both mistakes in the field constantly, running in opposite directions. Content-marketing teams build reputation and skip the infrastructure: the engine senses the brand is respected but has no clean entity to attach it to, so the citation never lands. SEO teams build the infrastructure and skip the reputation: the brand is cited as a source but never named as the answer, because no real person in any community AI reads has said so. Both have done real work. Neither has done the whole job. The gap between the two is where almost all the ground in AI visibility now sits.
This is why I built the second part of CLEO - the social side, the earned off-page work - and why I had to learn the platforms I had spent my life ignoring. Not to manufacture buzz; the engines see through that as easily as they see through a marketing blog. To do the patient work of being genuinely present where consensus forms, so that when the infrastructure makes a brand citable, there is something real for the engine to recommend.
The order matters, and it is the opposite of how most teams sequence it. Infrastructure first, reputation second. Build the reputation on top of an entity the engine cannot resolve, and the lift never arrives - six months on, the team concludes “Reddit didn’t work,” when the truth is the engine could not connect the Reddit mentions to a brand it recognised. Get the order right and the two compound. The engine sees the mention, recognises the entity, finds the structured page, and the recommendation holds.
That left one more thing I had to be honest with myself about - and with anyone paying us. Doing both halves of this work well still does not let me promise an outcome. That is the last piece.
More on how we think about this at regencleo.ai.
