Website Personalisation That Isn't Creepy

There is a version of website personalisation that makes people close the tab. You mention a brand of oven to a friend, then the oven repair company's homepage greets you by suburb and job type, and the whole thing feels like being followed home.

There is another version that nobody notices at all, because it just looks like a website that was paying attention. Someone clicks a Google ad for emergency hot water repairs at 9pm, lands on a page whose first line says you are taking urgent calls tonight, and books. No profile was built, no data was bought. The site simply used what the visitor had already told it by clicking that link at that hour.

Website personalisation done well sits entirely in the second category. It runs on signals the visitor knowingly handed over during this visit, and it changes the page in ways that make the next step easier rather than in ways that show off.

The signals you already have, and the line worth drawing#

Everything useful comes from five sources, and you own all of them.

The referring source tells you the intent behind the visit. A click from a Google Ads campaign about blocked drains carries a very different need to a click from your Instagram profile link, and you can read it from the URL parameters on the ad or the referrer header on the social link.

The page path tells you what the person is actually interested in right now. Somebody three levels deep in your commercial services section is not the same visitor as someone who landed on the homepage and has not moved.

Time and day tell you what the business can offer. A weekday at 10am and a Sunday at 8pm should not produce the same call to action, because your capacity to respond is genuinely different.

Returning visitor status, held in a first-party cookie, tells you whether the pitch has already been made. Somebody on their fourth visit does not need the explanation again.

Approximate location, from the browser or the IP, tells you which service area page and which phone number to show.

The line worth drawing is simple. Use what the person did on your site or knowingly gave you. Do not use data assembled about them elsewhere. That covers cross-site behavioural profiles and anything bought from a data broker. The practical test is whether the visitor would be surprised to learn where the information came from. Referral source and page history pass that test easily. A profile of their browsing across forty other sites does not, and in New Zealand it also drags you into obligations most small businesses have not thought about, which I have written up in the Privacy Act and AI.

Three website personalisation swaps worth making#

Personalisation gets complicated fast if you let it. These three cover most of the available gain for a service business, and each one is a single change to a single element.

Swap the hero line to match the entry point#

The headline on a landing page should answer the question the visitor arrived with. When the traffic all comes from one place, that is easy. When it comes from six places, a static headline has to be vague enough to cover them all, and vague headlines convert worse than specific ones.

So change it based on where they came from. Traffic arriving from a "burst pipe Auckland" search sees "Burst pipe? We can be there today." Traffic arriving from your Instagram profile link sees a line that continues whatever the post was about, because that visitor is browsing rather than panicking.

The offer underneath stays identical. You are matching language to the question, which is the same discipline that makes a good service page work in the first place. The section-by-section version of that is in writing a service page that actually gets enquiries.

Swap the call to action to match the hour#

The strongest personalisation signal for a service business is the clock, and almost nobody uses it.

During business hours, the primary button says "Call us now" and shows the number, because a phone conversation is your highest converting channel and somebody is there to answer.

Outside business hours, that same button becomes a liability. A person rings, gets voicemail, and rings the next business on the list. Replace it after hours with something that actually completes: booking the first available slot tomorrow, or starting a chat with an agent that can take the details and confirm a time. A large share of enquiries for home services and clinics arrive in the evening, because that is when people are free to deal with the thing that has been bothering them all week. What to do about after-hours enquiries goes through the options in more detail.

The same logic applies to public holidays and to the period between Christmas and mid-January, when half of New Zealand is at the beach and the other half is trying to find someone who will answer the phone.

Swap the proof to match the section#

Trust markers work best when they are close to the thing being sold. A commercial maintenance page should show your commercial work, not the same residential bathroom photo that sits on every other page.

This one requires no clever technology, only the discipline to write proof per section rather than sitewide. It is the cheapest of the three swaps and usually the one that lifts enquiries most, because a visitor deep in a service section is already close to a decision and specific proof is what closes the gap.

Where AI adds something and where it does not#

Rule-based swaps like the three above need no AI at all. They are conditions on a page, and any competent build can handle them.

AI earns its place when the range of possible visitor needs is too wide to write rules for. An agent that reads the visitor's actual question and assembles a relevant answer from your service information is doing something a rules engine cannot, because you would need a rule for every phrasing of every question. That is closer to on-demand personalisation than to swapping a headline, and it is one of the four capabilities that separate a smart site from a brochure site, covered in what makes a website smart.

The place AI does not add anything is deciding which swaps to make. Small business sites rarely have enough traffic for an algorithm to learn from. If you get a few hundred visitors a week, an automated optimisation tool will spend months failing to reach significance on anything, while a human looking at the same traffic could see in five minutes that the after-hours call button is the problem.

Keeping it honest#

Two rules keep personalisation on the right side of the line.

Never show a visitor information about themselves that they did not give you in this session. Naming their suburb because they typed it into a form is fine. Naming their suburb because their IP address suggested it reads as surveillance even though it is technically less invasive.

Never let personalisation change the substance. Adapting language to the visitor is fair. Showing different prices to different people based on inferred willingness to pay is not, and in a market the size of New Zealand somebody will notice and post about it. Price transparency is a conversion tool in its own right, and quietly varying it destroys the trust that made it work.

There is also a maintenance cost people underestimate. Every variant is another thing that can break, go stale, or contradict something else on the site. Three well-chosen swaps that you check quarterly beat fifteen you set up once and forget. If a rebuild is on the horizon, decide which variants survive it before the work starts rather than after, using the website redesign checklist.

What to do first#

Open your analytics and look at where your traffic actually enters the site and at what hour. Most service businesses find two or three dominant entry points and a large evening block that nobody has designed for.

Then make one change: the after-hours call to action. It takes an afternoon, it needs no personalisation platform, and it addresses the single largest group of visitors currently hitting a dead end on your site. Once that is running and you can see whether evening enquiries went up, pick the next swap.