Concept · 2026
Tidemark
A landing page concept for an analytics product that shows a business how AI assistants read, quote and recommend it.
- Role
- Product Design / Landing page
- Year
- 2026
- Stack
- Figma, Art direction, Copywriting
An essay that ends with a useful tool
Tidemark is a concept for the moment search changed: people now ask an assistant and take its word. Tidemark asks those assistants about your business every day and shows what they got right, what they got wrong, and which page caused it. I designed the landing page as an editorial story rather than a feature list.
The product is hard to picture, because the problem happens in a chat window the business never sees. So the page has to make an invisible problem visible before it can sell a fix. I open with an input and a promise, spend the middle explaining why answers go wrong with drawings, and only then show the dashboard, so the tool arrives as the answer to a problem the reader now recognises.
How it should feel
Calm, curious and a little literary. Watercolour landscapes and quiet paper tones slow the reader down, while line drawings and monospaced labels keep it precise.
It should feel like reading a thoughtful essay that happens to end with something you can use.
The split is deliberate. Everything human, such as headlines, paper tones and paintings, is set in warm cream and soft colour, while everything the machine produces is set in monospace labels, thin lines and one electric blue. Blue appears only on things that matter: the button, the wrong answers and the numbered markers. That is how a reader learns where to look without being told.
Up close
Your next customer asks an AI first
The visitor has just seen the hero and is asking why this matters to them, so the second screen makes the argument in three short sentences: people used to click, now they ask, and a wrong answer never reaches you. The headline sits left on plain paper, a single outlined button, Why answers go wrong, invites the next step, and the painting on the right does the emotional work. Over it, an isometric drawing of a page with a pop-up and a cookie banner stacked on top carries numbered mono labels, so the image is an argument and not decoration. The strip of assistant names at the bottom, with a small note that it works for any business, answers who and whether it applies to me.
Written for people, not assistants
Here the reader needs the problem broken into parts they can recognise on their own site. I used four equal columns, each numbered and tagged with a one-word mono label, then a thin line drawing, a short bold title and one sentence of explanation. The drawings are all drawn in the same isometric language, so the four feel like a set, and the tags run from the page itself to Unseen, which moves the reader from cause to consequence. Only the third column gets a blue marker, which is a small asymmetry that keeps the row from feeling like a table. The line drawings use no colour on purpose, so nothing here competes with the product later.
How an assistant reads your site
This is the section that explains what the product actually does, so I switched the metaphor from drawings to a wireframe. Two phone outlines on watercolour show a real page as an assistant reads it, with annotations in mono type: img, no text, pop-up with no close label, form with no labels. Numbered blue squares tie each fault to a place on the screen, and a dashed arrow with price not found links one page to the next. The copy panel sits to the right with the headline and four short capability lines, each with a small icon, so a reader can finish it in seconds. Dimension lines around the phone borrow from technical drawings and quietly signal rigour.
Hear the answer before your customer does
A visitor who believes the problem now wants proof that this is not a mock-up, so I show one complete exchange as a diagram. On the left a phone shows ChatGPT answering a question about stock and returns; on the right Tidemark checks the same question against the site. The mono tags, Asked, Check and Answer, make the sequence readable, and the answer carries its two sources, /trail-runner-2 and /returns, so the reader sees how a claim is traced to a page. Three numbered points underneath, Every answer checked, Sources you can see, Every answer in your log, restate the benefit in the order a business owner would worry about it. The line saying nothing on your site changes removes the first objection.
See what AI says about you
The product itself finally appears, so I kept it modest and readable. The left column names three things the dashboard answers, who answered, what it was asked and where it went wrong, so a reader knows what to look for before looking. The screen shows four numbers at the top, a single stacked bar chart that splits correct and wrong in pale and solid blue, and a table with the assistant, the question and the page. Wrong answers are the only items filled with blue, so the problem rows are found at a glance without reading. The phone version beside it shows that the same story survives a smaller screen. The closing button, See AI mentions, is the same outlined style as the earlier buttons.
See the full pageThe whole design, top to bottom, in full quality.
Looking back
Pairing fine art with product UI was a risk. It works because both ask the reader to slow down and look closely, which is exactly what this product asks of a business. It is a concept, with illustrative numbers, so I would not claim results. The next step would be to tighten the long middle of the page and check that a first-time visitor reaches the dashboard without losing the thread.