BookModels: a 2 month rebuild that 5x’d search clicks and lifted daily revenue 715%

BookModels is a two sided marketplace. Models list profiles on one side, advertisers book them on the other, for events, shoots, conventions, brand work, and product demos. The product worked, in the narrow sense that it was online. Everything underneath it was a problem.

It ran on PHP and MySQL inside a GoDaddy cPanel account, and it had not been touched in a long time. Pages were slow. On mobile, most of the site scored in the low 50s for Core Web Vitals, and the pages people actually convert on, login, hire, the article hub, dropped into the 40s and the 30s.

The business model had the same shape as the codebase. One lifetime offer. A model paid once and was listed forever. A lifetime price feels generous to the customer and looks clever on a landing page. It is also a slow way to run out of money on the people who like your product the most, because the relationship turns into pure cost the day after they pay.

So the brief was not a fresh coat of paint. It was a full rebuild, with the search rankings kept intact, and a revenue model that grows instead of caps.

The marketing site, the model and advertiser app, and the admin that runs it, all rebuilt.

The marketing site runs on Astro. It is static and fast, and the home page ships almost no JavaScript to the browser, which is most of why it loads the way it does. The app runs on Next.js. Data, auth, and storage sit on Supabase. The whole thing is hosted on Vercel, with Cloudflare handling DNS and image storage. None of that stack is exotic. The point of naming it is that boring, well chosen infrastructure is what lets a two person job hold up under 50,000 profile pages.

The accepted wisdom on a site migration is that you brace for a traffic drop and claw it back over the next three months. We did not want to spend a quarter recovering ground the old site already held.

So we kept every URL slug that was already ranking, exactly as it was, and set 301 redirects only where the structure genuinely changed (the old cities path now canonicalizes to /locations). Search engines saw the same addresses pointing at faster, better pages, not a new site they had to relearn.

Clicks went up after launch instead of down. You can see the line.

The growth is not one or two hero pages. It is a large, genuinely useful long tail built from real data:

That is tens of thousands of pages, and they are organized so search engines can crawl them without choking. The index is split across a primary sitemap plus content, profile, and photo sub sitemaps, with a robots file that points at the right places. Big surface, clean structure.

Mobile Core Web Vitals went from the low 50s to 99 and 100 across the site. The home page passes Core Web Vitals on real Chrome user data, on both mobile and desktop, with Lighthouse Performance and SEO each at 100.

A couple pages were already light on the old site and scored well. The pages that mattered most for converting visitors were not, and those are the ones that moved the furthest.

PageSpeed now scores the site 2 of 2 on agentic browsing. That is a 2026 way of saying AI crawlers and assistants can read the pages and act on them. As AI search starts sending real referral traffic, a site that is already legible to it keeps compounding while everyone else scrambles to retrofit.

A two sided marketplace has two funnels: getting models to list, and getting advertisers to book. You cannot run paid acquisition or lifecycle email against either one if you cannot see which actions matter. The old site could not. There was no usable measurement on it at all.

We wired the same event model we run on our own products. Every event belongs to exactly one of three categories, and the category decides where it fires from.

On top of that taxonomy: GA4 for product analytics, Google Ads for paid attribution that ties back to revenue rather than to clicks, and Microsoft Clarity for session replay and heatmaps, so we can watch where real people get stuck instead of guessing.

All of it runs behind a consent gate that handles the EU (opt in), the US (opt out), and a worldwide default, so analytics and ads only fire on the legal basis each visitor allows. Compliance and measurement are not at odds when the consent logic is built in from the start rather than bolted on.

The lifetime offer is gone. In its place, recurring revenue on the model side and consumable tokens on both sides.

Here is the part that pays for itself. A lifetime price collects once, and after that the most engaged users on the platform are the most expensive ones to keep serving. Recurring revenue plus usage based tokens flips that: the more value a model gets from being seen and the more an advertiser gets from booking, the more the platform earns, every month, from both sides at once.

Average daily revenue rose 715% after the change.

Most feedback dies in an inbox. We put a Feedback button in the corner of every dashboard, and what a user types there does not drop into a support queue nobody reads. It opens a ticket on the dev team’s Jira board, tagged by type, in the same To Do, In Progress, Done flow the engineers already work from.

Then the loop closes, which is the part most products skip. When a suggestion is acknowledged, when a bug is fixed, when a requested feature ships, the person who asked for it hears back, through in product messaging plus email or Telegram, whichever they have connected.

A roadmap built from what users actually ask for beats one built from what a team assumes they want, and telling someone you shipped their idea is some of the cheapest retention there is. The board fills with the real stuff: preview a profile before it goes live, see who you rank against in a category, fix a city that got assigned wrong. Some of it is just praise. All of it reaches the people who can act on it.

The old site read like a directory from 2014, right down to a blog post titled “Top 10 Butt Models.” The new one is editorial. Real typography, a clear position as the inclusive modeling platform, and photography that looks like a brand instead of a stock library.

You cannot put a single clean number on brand lift. You can see it in the before and after, and you can see it in who replies to a booking request now versus who did before.