Updated September 24, 2026 · By Emiliano Ubaldi
Good personalization should feel like a business paying attention. It might help a returning visitor find a saved vehicle or show a first-time researcher a clear comparison. It should not leave someone wondering how a website knows so much about them.
Prediction and generation play different roles
Predictive AI estimates a likely outcome or next action from available signals. Generative AI can produce content for that action. Adobe’s predictive AI overview describes uses such as propensity modelling, while its decisioning tools combine profile and contextual information to select experiences.
Those capabilities do not mean every business needs a unique page for every person. The useful question is much narrower: which decision is currently difficult for a customer, and what information would make it easier?
Start with one moment of friction
Imagine a visitor returning to research SUVs. If the site has an appropriate basis to remember their preference, a shortcut to relevant available vehicles may be helpful. A new visitor may benefit more from a comparison guide. These are illustrative experiences, not evidence that a model can reliably infer someone’s budget, family situation or readiness to buy.
I would begin with one journey, a small set of trustworthy signals and a clear fallback for visitors with limited data. Explicit preferences often provide a stronger starting point than elaborate assumptions. Keep prices, stock details and offer conditions connected to verified information; generated wording must not invent them.
Measure whether the experience earns its place
Set a success criterion before launching. For a lead-generation website, that might be a higher proportion of relevant enquiries or fewer abandoned booking attempts. Compare the personalized experience with a standard version using a controlled test where practical.
Check more than clicks. Extra engagement can come from confusion, and a shorter journey is not always better if visitors miss important information. Review lead quality, usability and complaints alongside conversion figures.
Data collection should be understandable, proportionate and consistent with the visitor’s choices. Limit access and retention, and give the team a simple way to switch back to the default experience if a recommendation becomes inaccurate.
My preference is to earn complexity one useful improvement at a time. Let’s identify where a more relevant experience could help your customers.
Illustration generated with AI for IM Advise.