My home page leads with Work. That suits someone deciding whether to hire me. It suits someone who arrived from an essay much less: they have to scroll past five case studies to find more of what they came for. I would like a lighter, Notebook-first journey for people exploring and a Work-first one for people hiring.
The tempting next step is to guess which visitor is which and rearrange the page. This post is about why I have not done that yet, and what I am doing first.
A browser does not know who is hiring
Every browser hands a page plenty of context without asking. None of it says what the person wants.
| Signal | What it is good for | What it cannot tell me |
|---|---|---|
| Screen size, orientation, pointer | Layout and controls | Intent |
| Browser and its capabilities | Compatibility and performance | Profession or budget |
| Time and time zone | Presentation | Real location or availability |
| Preferred language | A language default | Nationality |
| Reduced motion, colour scheme | Respecting preferences | Anything personal |
| The link they arrived by | Interest in that content | Who they are: links get shared |
A laptop at 11 a.m. on a weekday is not a client, and a phone at midnight is not a casual reader. Treating them that way would be guessing with extra steps, and storing those guesses to use on the next visit would need consent I have no reason to ask for.
Device / Time / Language / None of them is intent
What real personalisation runs on
I read the documented cases I could find. They are accounts by vendors and their partners, not independent audits, but they agree on one thing.
Saks (opens in a new tab) changes its home page by purchase-intent tiers built from first-party data, and those tiers can change within a visit. Jacamo (opens in a new tab) features the brands and categories a visitor showed interest in before, on their return. Livestorm and Dropbox DocSend (opens in a new tab) change logos, copy and proof by company size and industry, which they buy from a data-enrichment service, plus whether the visitor is already a customer.
Every one of them runs on history, bought company data or a known customer. None shows that a first visit can be classified reliably from the browser alone. A one-person studio has none of those inputs, and I do not want the ones I would have to buy.
Purchase history / Company data / Known customers
Measuring journeys without knowing anyone
So the first step is to look at what visitors actually do. The analytics I added are deliberately narrow:
- No cookies, no saved identity. A random visit ID lives only in memory. A reload or a new tab starts a new one, and so do thirty minutes without activity. Counts are visits, not people.
- Meaningful events only. A page shown; a home section on screen for a second; a case study or note deliberately opened; a note read to its end with at least thirty seconds in the foreground; a contact action. No automatic capture of every click, no text from the page.
- Declared context, kept separate. A link can say who is visiting (me, a tester) and what they came for (work, reading, play). Those labels go on the events and never decide anything about the person.
- Recordings are separate and disclosed. Some visits can be recorded as masked session replays under rules set out on the site's privacy page, which also lets a visitor refuse them. Replays help me see where a page confuses people; they are not how journeys are counted.
From those, a few reports answer the question directly: the paths visitors take from the home page, how many who see Work open a case study, and what people do before they get in touch.
No cookies / A visit, not a person / Thirty minutes
The blind spots I know about
The page shapes the data. Work comes first, so more people see it, so more people open a case study. That says nothing yet about whether Notebook-first would do better. Comparing groups needs the paths and what each visitor was shown, not just the totals.
Engagement is not reading. Reaching the end of a note after thirty seconds is a better signal than a page view, but it is not proof. Contact is not a sale. And there will always be visitors who fit no group, so the reports keep an "unknown" group instead of forcing everyone into one.
Exposure / Engagement / An unknown group
What happens after
I have no results yet. When there is enough real traffic, the options in order of how little they risk:
- Keep one concise home page, with Work and Notebook highlights both near the top.
- Use deliberate entry links: a Work-first route for proposals, a Notebook-first route for essays.
- Let a visitor's first real choice, opening a case study or a note, shape what is suggested next.
- Reorder the page itself only if the evidence supports it, and never while someone is scrolling.
Every destination stays one link away for everyone. I will write the follow-up when the data says something, including if it says the current page is fine.
Nothing moves under a reader
Measure the paths before you redraw them.