What is a smart website? Pages that adapt to the ad that brought you here

The afternoon experiment

Here is an experiment you can run this afternoon. Open your ad accounts, pick your five best performing ads, and click each one as if you were a stranger. Then look at the page you land on, and ask one question: does this page continue the conversation the ad started?

Very often it does not. The ad promised a fast quote for a specific problem, and the page is the homepage, with a slider, a mission statement and a menu with eleven entries. The ad spoke to a wholesaler, and the page speaks to consumers. The ad was in the informal register that works on LinkedIn, and the page reads like a brochure from 2014.

You paid for the click. The page threw it away.

At some point the obvious question is why the website cannot do this itself. That question is what I call a smart website, and this article is the long version of what I mean by it, how it behaves, what it deliberately refuses to do, and where it stands today.

One page for everyone is a compromise nobody chose

Think about how a typical company website comes to exist. Somebody writes a brief. An agency or a developer builds a set of pages. The copy on each page has to work for every possible visitor at once: the person who has never heard of you, the person comparing you with two competitors, the existing customer looking for the support number, the journalist, the job applicant, the buyer from a trade fair and the buyer from a Google search. So the copy is written to offend nobody and to speak to nobody in particular. That is not a failure of the copywriter. It is what the brief asked for.

Then marketing starts running ads. And ads, if they are any good, are the opposite. A good ad is specific. It names one problem, one audience, one promise. The whole discipline of paid media is about narrowing, about finding the message that makes one segment stop scrolling. So you end up with a funnel that has a very sharp top and a very blunt landing.

The usual answer has been the landing page: a separate page per campaign, built outside the main site, often in a page builder that is a second content system with its own login, its own hosting bill and its own way of breaking. Landing pages work. But look at what they cost. Every campaign needs a new one. They drift out of step with the main site's design and claims. They are orphans in search. They usually strip out the navigation, so the visitor who wants to know more has nowhere to go. And the moment the campaign ends, they rot.

What I wanted was the benefit of a landing page, a page that continues the ad, without the tax of maintaining a parallel website. That means the main site itself has to change what it says depending on why you arrived. That is the whole idea.

What message match is, and why it matters

Message match is a plain concept: the page a visitor lands on should repeat, in words and in substance, what the ad promised. Same headline idea, same offer, same tone, same audience. When it does, the visitor registers "right place" in under a second and reads on. When it does not, they get the small jolt of "hang on, is this the same thing?", and that is exactly the moment people leave.

Google builds the same idea into its own diagnostics. Quality Score in Google Ads is made of three components: expected click through rate, ad relevance and landing page experience, which Google describes as how relevant and useful your landing page is to people who click your ad. Google is explicit that Quality Score is a diagnostic tool, not an input in the auction, but the components it diagnoses are the ones the auction cares about.¹ A page that continues the ad is the page Google is asking you to build.

I wrote about the wider framework in the message market model fit piece: the message has to fit the market and the model before you spend a pound. A smart website is the mechanism that keeps the message fitting the market on the page, not just in the ad.

The model: one canonical page, shaped versions of it

Here is the shape of a smart website, from the outside in.

There is one canonical version of every page. That is the version search engines index, the version a visitor gets when they type your domain into a browser, and the version anyone gets when the site has no reason to show them anything else. It is a proper page, complete and well written, and it stands on its own. Everything else depends on that.

Then, for the pages that matter, there are shaped versions. A shaped version is the same page with certain parts changed: the headline, the opening paragraph, the image, the proof, the call to action, sometimes the order of the sections. It exists because a specific ad, a specific audience or a specific stage in the journey needs the page to say something more precise than the canonical version can.

The site decides which version to show from two things: why you arrived, and what it is allowed to know about you. The campaign you clicked connects the page to the ad, so the promise carries on and fewer people bounce. What the site may lawfully know about you, such as an interest you agreed to have remembered or where you are in the journey, moves you further along that journey.

Five dimensions, and the sixth that keeps them honest

I build smart websites in five dimensions, with a sixth layer that tests all of them. I call the result a Facet. The names are short on purpose, because each one is something a visitor feels and an owner can measure.

Story. The page stops being a brochure that lists things and becomes a journey: each section sits where a visitor makes a decision, from the problem they arrived with, through what changes, the proof and the plan, to the next step. I have written about the StoryBrand structure behind this.

Signal. Every section reports back, so the numbers you look at are the numbers that happened, not a sample with holes in it where a banner was declined or a tracker was blocked. Counting that something happened is a different thing from following a person, and the difference matters. More on that in the server side tracking guide.

Stage. The site notices where a visitor is: just looking, weighing it up, or ready to talk.

Shape. The page continues the promise the visitor came in with, and speaks to where they are.

Share. When anyone shares the page, on a network, in a message or in an ad, the preview image and text match the message it was shared with, so people land on a page that looks like what they clicked.

Prove. Every one of those layers can be tested properly, against real enquiries, so the winner is data rather than a hunch.

Why you arrived drives the page. Consent decides whether the site may remember it.

I will not go into how each layer is engineered. That is my job, and it changes as I improve it. What matters for an owner is what each layer does and what it refuses to do, so the rest of this article stays on that.

Why you arrived: the link is the contract

Every ad platform lets you add parameters to the link a click goes to. Most people know them as UTM tags and use them only so that analytics can report on which campaign a visit came from. I wrote a whole guide to using them properly.

A smart website reads those same parameters as a request. The campaign and a small amount of extra detail in the link say, in effect: this visitor clicked the ad that promised fast quotes for warehouse racking to logistics managers. If a shaped version exists for that promise, the visitor sees it. If not, they see the canonical page. No guesswork, no inference, no third party data. The ad and the page are written together, as a pair, and the link is the contract between them.

This is also why the first visit needs no stored data at all. The visitor brought the information with them, in the link they chose to click.

Remembering, but only with permission

The first visit is the easy case. The second is the interesting one.

A logistics manager clicks your LinkedIn ad on Tuesday, reads the racking page, and leaves. On Thursday they type your domain into the browser and land on the homepage. What should it say? The lazy answer is "the same as for everyone". The useful answer is "the version that continues Tuesday's conversation": the warehouse case study instead of the generic one, the quote form they saw before instead of the newsletter box.

To do that the site has to remember that this visitor was interested in racking, and here I draw a hard line. The site remembers only if the visitor has agreed, and it remembers in a first party cookie that holds the interest and the version they saw, nothing else. Not a user ID. Not a fingerprint. Not a profile stitched together from other sites. Nothing is stored before they agree, and it is gone when they withdraw or when the cookie expires. Stage works the same way: carrying what someone looked at into their next visit needs their agreement.

This is not only my preference. Under regulation 6 of the Privacy and Electronic Communications Regulations, a site must not store information on a visitor's device, or read information already there, except with consent or under the narrow exceptions in Schedule A1.² Those rules are enforced by the Information Commission, which replaced the ICO on 30 September 2026.³ In the UK some of the new exceptions are a judgement call, and I treat them as a risk decision for the client rather than a loophole. Asking is the safer and the more honest route.

If the visitor has not agreed, the site shows them the page shaped by the link they clicked, or the canonical page. It does not degrade, it does not nag, and it does not try to work out who they are from their IP address or screen size. Consent unlocks continuity, and the visitor who says no still gets the full site. I went deeper into where the line between helpful and creepy runs in privacy first growth hacking. The short version: the line is whether the visitor could predict what you know about them. The ad they clicked, remembered with permission, is predictable. Anything else is not.

Testing is the architecture, not a plugin

Once a page can show different content for different reasons, A/B testing stops being a separate tool you bolt on. It is the same mechanism with one more input.

Every shaped version can have alternatives. The racking page might have two headlines in a test. A visitor who matches the racking version is assigned one of them, the assignment stays the same for that visitor, and the enquiry is counted against the exact combination they saw. There is no script from a testing vendor that makes the page flicker, and no separate calculator in another tab.

The reporting follows from that. For every promise you can see how many people arrived, how many enquired and which alternative won. The point I keep making in the guide to digital marketing testing methods is that most tests are declared winners far too early. When the test is part of the site, the site can refuse to call a winner until the numbers can carry it, instead of relying on a marketer under pressure to hold their nerve.

The consequence for the business is simple. The site gets better every month, per audience, on evidence, without anyone redesigning anything.

Where the content lives, and why there is no CMS

This surprises people, so let me be precise. A smart website has no content management system in the WordPress sense. No admin with a page tree, no visual builder, no plugin directory.

The content lives as structured pieces. A headline is a piece. An opening paragraph is a piece. A proof block with a quote and a name is a piece. Pages are compositions of pieces, and shaped versions swap some of them. A page builder stores blobs of layout, and you cannot sensibly ask a blob for "the racking version of this page". You can ask that of structured pieces.

So how does the owner change the copy? Through the client portal, in plain language. You describe the change, a draft appears in your brand voice, you approve or edit it, and it goes live. Nobody logs into a CMS, and nobody waits a week for a two line change. Every change is recorded, with who approved it and when.

The boundary is firm. AI drafts. A person approves. Nothing on a smart website is published by a model without a human pressing the button. The value of AI here is that the tedious part, writing seven versions of the same headline for seven audiences, takes minutes instead of a day. The judgement about which one goes live stays with you.

Search engines, canonical tags and the cloaking question

The first question any competent SEO asks is: is that not cloaking? It is the right question. Google defines cloaking as presenting different content to users and search engines with the intent to manipulate rankings and mislead users, and treats it as spam.⁴

The answer is no, for structural reasons rather than hope.

First, the canonical page is a real page and it is what search engines see. A crawler arrives without campaign parameters and without a consent cookie, so it gets the canonical version, exactly as a first time visitor typing the address would. There is no special treatment for bots in either direction.

Second, shaped versions change emphasis, tone, order and examples, not what the page is about. The product, the service, the claims and the prices stay the same. A shaped version of the racking page is still the racking page. Google's own guidance on testing says the same thing: point alternate URLs at the original with a rel="canonical" link, never show crawlers something different from what users see, and note that small changes such as a headline or a call to action often have little or no impact on the snippet or ranking.⁵

The effect is that the canonical site tends to get stronger in search, because the campaign effort that used to go into orphaned landing pages now goes into pages that are indexed. In the piece on SEO, AIO, GEO and AEO I argue that clear, well structured canonical pages are exactly what AI answers want to cite.

What a smart website refuses to do

Personalisation has a bad reputation, and it earned it. So it is worth saying the limits out loud.

It does not fingerprint. It does not combine screen size, fonts, time zone and a dozen other signals into a shadow identifier that recognises you without a cookie. European regulators read the device rules broadly: the EDPB's guidelines on the technical scope of Article 5(3) of the ePrivacy Directive cover information a script gathers from the device, not only cookies.⁶ In practice that means the same consent question applies to fingerprinting, and the honest answer is not to build it.

It does not guess beyond what it may know. No inferring age, gender or income from behaviour and changing the page on the guess. Every mechanism here keys on two lawful inputs: the campaign you clicked, and what the site is allowed to know about you, which for anything remembered means your agreement.

It does not put personal data into the shaping layer. Your name and email from a form go where they belong, into a CRM or an order system with its own purpose and its own consent. The part of the site that shapes pages never sees them and does not need them.

It does not buy or borrow data. No third party audience segments, no enrichment services, no "this IP belongs to a company in Leeds" tricks. If you want the site to speak to logistics managers, you run an ad for logistics managers and the site continues it. The audience definition lives in the ad platform, under that platform's rules, and the site respects the hand off.

These limits are not reluctant. They are why the approach works in Europe at all, and why it will keep working when the next round of regulation lands.

A worked example: a garden furniture shop in Bristol

Let me make it concrete with an invented business. Harbourside Garden Co., at harbourside.example, is a Bristol shop selling garden furniture online: a few hundred products, a design led brand, customers across the UK. Nobody real, but the situation is common.

The canonical homepage is what it should be: a seasonal hero, the main categories, reviews, a newsletter box. It ranks, it converts cold traffic reasonably, and there is no reason to change it.

Now it is April and the shop runs three campaigns. On Meta, a video about a modular sofa set. On Google Search, ads on "weatherproof garden lounge". On Pinterest, a campaign around small balcony furniture.

On a conventional site all three land on the homepage or a category page, and the balcony visitor scrolls past a hero of a lounge that would never fit on their balcony. On a smart website each campaign has its shaped version of the landing page. The Meta visitor sees the modular sofa as the hero, with the promise from the video as the headline. The search visitor sees "weatherproof" in the first line, with the weather guarantee as the proof, because that is what they searched for. The Pinterest visitor sees balcony sets first.

Each visitor who agrees to be remembered is remembered as "modular", "weatherproof" or "balcony". When they come back on their own in May, the homepage continues from there. Those who do not agree get the page their link shaped on the first visit, and the canonical homepage after that. Every shaped version runs two headlines in a test, so by June the shop knows which line sells sofas to the Meta audience, without a single meeting about it.

What it took: three shaped versions of one landing page, written in the portal with the campaign links set up at the same time. What it replaced: three separate landing pages in a builder, a testing vendor's script, and a monthly argument about which hero image to use.

A worked example: an engineering consultancy in Manchester

Now the other end, again invented. Pennine Structures, at pennine.example, is a fifteen person engineering consultancy in Manchester doing structural surveys and compliance reports for commercial property, with a website that lists eight services and speaks to everyone.

It runs LinkedIn ads to two audiences: facility managers at property companies, and architects at mid sized practices. The facility manager ad is about avoiding a failed inspection. The architect ad is about fast turnaround on load calculations during a design phase.

Conventionally, both land on the surveys page, which lists everything in the order the founder thought of it, and the facility manager has to work out which of the eight things applies to them.

On a smart website, the facility manager sees the surveys page shaped around the inspection: the outcome as the headline, the most relevant certification up front, a case from a property company, and a call to action to book a pre inspection walk through rather than "contact us". The architect sees the same page shaped around turnaround time, with a call to action to send drawings. Same service. Same page in Google's eyes. A completely different first thirty seconds for the person reading it.

Remembering matters even more here, because B2B buying takes weeks. The facility manager comes back three times over a month, from a saved link, from a colleague's forward, from a search for the company name. With their agreement, each visit continues the inspection conversation instead of resetting to the founder's eight services. When they finally enquire, the enquiry arrives tagged with the promise they came in on, so whoever calls back knows why they came.

Why Rails, and why not a page builder

People ask whether all this could be a WordPress plugin or an app on a hosted site builder. Some of it could. I have chosen not to, for the reasons I gave in my comparison of web frameworks.

A smart website is an application, not a collection of documents. It has pages, pieces, versions, promises, tests and enquiries, logic that runs on every request, and a portal with roles, approvals and an audit trail. Ruby on Rails is built for exactly that: a mature framework with proper testing and a data layer that expresses these relationships cleanly. A page builder is built to let people arrange boxes, and bolting application behaviour onto one ends up fighting the tool.

There is a security argument too. A site whose behaviour depends on thirty plugins from thirty authors depends on thirty people you have never met. A small, tested Rails codebase can be checked on every release. And there is an economic argument: when the design is a layer over structured content, it can change without rebuilding the content, which is where much of the money spent on periodic website rebuilds goes.

Where it stands today

I would rather tell you what exists than sell you a brochure.

keferboeck.com is the first smart website. The homepage, the about page and the Smart Websites service page shape themselves to the campaign a visitor arrived from and to what the site is allowed to know about them, and the previews follow the message when they are shared. Remembering is consent based, first party, and nothing is stored before agreement. The editing loop in the client portal, describe, draft, approve, live, is how these sites are run, and this site is the first one it is being built for. The portal itself already exists and already handles documents, approvals and an audit trail for clients.

So if you want one now, what you get is an engineered Rails site with the five dimensions and testing built in, and the editing loop as it lands. What you will not get is a promise that every idea on this page is a button you can press today.

The bit I keep coming back to

Strip away the mechanics and the idea is old fashioned. A good shopkeeper talks to each customer about the thing they came in for. They do not recite the whole inventory to everyone who walks through the door, and they do not follow people down the street to find out where they live. They remember the regulars, because the regulars want to be remembered, and they treat strangers as strangers until the strangers say otherwise.

Websites lost that when they became documents. Ads never lost it, which is why ads are specific and websites are vague, and the gap between them costs everyone money. A smart website is simply the website behaving like the shopkeeper again: continuing the conversation the visitor started, remembering with permission, and improving from what actually works rather than from what someone thinks looks nice.

If your site does not do that yet, the free workshop is where we start: your ads, your pages and the gap between them. Then the first two weeks of work are on me.

Footnotes

1. Google, About Quality Score, Google Ads Help, accessed 10 November 2026, https://support.google.com/google-ads/answer/6167118?hl=en-GB

2. The Privacy and Electronic Communications (EC Directive) Regulations 2003, regulation 6 (as amended), legislation.gov.uk, accessed 10 November 2026, https://www.legislation.gov.uk/uksi/2003/2426/regulation/6

3. Data (Use and Access) Act 2025, section 118, in force 30 September 2026, legislation.gov.uk, https://www.legislation.gov.uk/ukpga/2025/18/section/118

4. Google, Spam policies for Google web search, cloaking, Google Search Central, accessed 10 November 2026, https://developers.google.com/search/docs/essentials/spam-policies

5. Google, Minimize A/B testing impact in Google Search, Google Search Central, accessed 10 November 2026, https://developers.google.com/search/docs/crawling-indexing/website-testing

6. European Data Protection Board, Guidelines 2/2023 on the technical scope of Art. 5(3) of the ePrivacy Directive, version 2.0, https://www.edpb.europa.eu/documents/guideline/guidelines-22023-on-technical-scope-of-art-53-of-eprivacy-directive_en