There is a quiet crisis at the heart of digital experience, and it is a crisis of abundance.
The amount of data platforms hold about user behaviour has grown beyond anything a human team could ever curate. At the same time, the universe of things a user could be shown — products, videos, articles, listings, results — has exploded into the millions. Search, in its classical form, was built for a world where the challenge was finding the needle.
We now live in a world where the challenge is that everything looks like a needle.
Think about what actually happens when someone searches on a large marketplace or content platform today. A generic query can legitimately match tens of thousands of items. Ranking them by simple relevance produces a wall of nearly identical options, and walls of options paralyse people. The paradox is well known: beyond a certain point, more choice reduces satisfaction and reduces action. The user does not want the complete set of matches.
They want the three that are right for them.
This is why I believe personalization is shifting from a feature of search to the substance of it. The query is becoming only one signal among many. Who is asking, what they have done before, what context they are in, what people like them chose in similar moments — these signals increasingly do more work than the keywords themselves. Two people typing identical words should not, and increasingly do not, see the same results. Search is quietly turning from a retrieval system into a decision-making proxy.
AI is what makes this viable at scale. No rules engine written by humans can map millions of users against millions of items across constantly shifting context. Learned models can, because they compress behavioural patterns into representations that generalise.
The interesting consequence is that the quality of a search experience now depends less on the algorithm — which is increasingly commoditised — and more on the feedback loop feeding it. Can the platform capture signals quickly, clean them, and push learning back into the experience within hours rather than quarters? That loop, not the model, is the competitive asset.
But I want to flag two tensions that I think will define the next phase. The first is the filter problem. When a system chooses for you, it also chooses what you never see. Personalization that optimises purely for engagement narrows people's worlds — showing them more of what they already liked, starving them of discovery. The platforms that handle this well will deliberately inject novelty and treat serendipity as a design goal, not an inefficiency. Most will not, because narrowness converts better in the short term.
The second tension is trust. Personalization runs on data, and users are growing more conscious of the trade. My sense is that tolerance depends entirely on perceived benefit: people accept deep personalization when it visibly saves them time and effort, and resent it when it feels like surveillance dressed as service. The line is not how much data you use. It is whether the user feels the data is working for them or on them.
There is also a coming structural shift worth watching: conversational interfaces are starting to replace result lists altogether. When a user asks an assistant rather than querying an index, personalization stops being a ranking adjustment and becomes the entire interaction. The assistant that knows you well enough to answer with one good option, rather than ten plausible ones, wins.
The question I keep turning over is this: when systems choose for us this effectively, who audits the chooser? Because the better personalization gets, the less we will notice everything it decided we should never see.
The above reflects my personal views only and is intended for informational and discussion purposes. It does not represent the position of any employer or organisation.