How BrandViz.AI tripled on-site conversion for NightCafe Studio by rebuilding the system that writes its pages

90 days, by the numbers

NightCafe had 52,000 programmatic landing pages, four years of organic decline, and no capacity to write anything. So BrandViz.AI didn’t write. The sprint went on diagnosis first, then on rebuilding the system that generates NightCafe’s tool pages. Five generations in the table below means the share of visitors who go on to make five images, which is where a visitor stops being traffic and starts being a user.

Google Search Console, Bing Webmaster Tools and Amplitude, 2 June to 25 August 2026.

The AI rows count impressions in Google’s AI surfaces only, and the Bing row is monthly rather than weekly. Part of the work was deliberately removing pages, so the click and conversion figures are the cleaner read.

Who is NightCafe Studio?

NightCafe Studio is one of the largest AI art platforms in the world: more than 30 million registered users, around 10 million images generated a month, and 30-plus AI models in one place, from Flux to Google Imagen to Stable Diffusion. It was founded in Brisbane in 2019, is bootstrapped, and is run by a small team.

A prompt box and what comes out of it. Every tool page on the site exists to start that loop, which is why the page a visitor lands on decides whether they generate anything at all.

It is also a platform under pressure. Midjourney and OpenAI moved into the category, NightCafe’s organic performance had been sliding for roughly four years, and paid acquisition wasn’t scaling profitably. Organic was the channel that had to work.

What was NightCafe’s organic position before the sprint?

NightCafe had no AI visibility measurement of any kind. It had never used Bing Webmaster Tools, had no read on AI citations, and no conversion analysis by landing page type, so none of its 52,000 pages could be ranked by what it was actually worth.

Two things defined the starting point. The first was scale that resists conventional work: on top of those 52,000 pages sit one to two million indexable user creation pages. There is no version of “write 45 articles” that touches a site that size.

The second was a known defect nobody had the capacity to chase. A bug in the original build had left some tool pages carrying content that didn’t match their own titles, so a page promising one thing described another.

The diagnosis that changed the plan

The first two weeks were analysis, not production. Three findings redirected the sprint. NightCafe’s blog drove under 1% of its clicks, so content volume was never the thing to pull. Its conversion surface is tool and landing pages, where purchase intent already arrives. And the pages that mattered were already built.

A fourth constraint turned out to be an advantage. NightCafe wouldn’t hand over repository access, because the whole application lives in one repository. Every fix therefore had to be written as a specification NightCafe’s own agents could execute, which forced reusability: nothing delivered was a one-off edit. The sprint became a rebuild of the page generation system rather than a batch of pages.

What did BrandViz.AI build for NightCafe?

NightCafe’s tool pages are written by software. So BrandViz.AI left the pages alone, rebuilt the software that writes them, then ranked which pages to send through it first. Four pieces of work, and one of them improved every page at once.

What did the prompt and search data reveal?

Crossing what people searched for against what NightCafe’s users actually made produced two findings, and both pointed away from writing anything new.

The biggest wins were already on the site. Google showed the AI logo generator 6,413 times in a month and it got four clicks, because it sat at position 14, while NightCafe’s own users wrote 4,555 logo prompts a week. T-shirt design, colouring book, tattoo and hairstyle looked the same. Every one of those pages existed already; the problem was where they ranked.

Identity beats technique. Pages about who someone is convert several times better than pages about how something looks: birth-flower bouquet turns 8.93% of its Google impressions into clicks, animal hybrid 6.48%, on terms nobody types casually. The clearest gap was fursona: 1,040 prompts a week from NightCafe’s own users, and no page at all.

What were NightCafe’s results after 90 days?

Conversion tripled, which is the number that touches economics

The share of visitors who go on to make five images moved from around 1.5% to 4.5–5%. That is the line between traffic and users, and it is what acquisition economics are built on. Every other number here is upstream of it.

Clicks up 60%, click-through up 0.8 points

Tool page clicks grew about 60%, from roughly 5,000 to 8,000 a week, and click-through rate rose from 3.4% to 4.2%. Impressions can climb on their own; a click-through rate climbing with them means the pages now match what people wanted. All of that happened while the site got smaller. Half the programmatic pages were removed for pulling volume without converting, taking the footprint from 52,000 to about 25,000.

AI surfaces close to doubled

Tool pages surfaced in Google’s AI results went from roughly 20,000 to 40,000 a week, and total AI impressions rose 76%. On Bing and Copilot, monthly AI impressions grew by about 20,000, from roughly 90,000 to 110,000, so the lift wasn’t Google-specific.

The superhero generator was still climbing six weeks later

The superhero generator moved from position ~12 to ~7 and was still climbing six weeks after it was touched, and the graffiti generator became NightCafe’s top tool page by AI impressions. That is the loop doing its work: better-matched pages hold people on-site longer, longer sessions and higher click-through push ranking up, and higher ranking brings more of the right traffic.

Recovery takes two shapes here. The page at the top of this study is the first: flat for two months, then near-vertical. Two more tool pages show the second, climbing from the start of the window and still climbing at the end of it, which is the pattern that compounds.

Google Search Console, daily, 31 May to 29 August 2026, covering the sprint and the weeks after it. Each chart above is a single tool page rather than the site, so the totals are page-level and are not comparable with the site-wide figures in the table near the top of this study. Search Console reports Google only, which is why the AI surfaces measured elsewhere in this study do not appear here.

Safety findings outside the scope

Checking which user creation pages Google was actually sending traffic to turned up around 120 that should never have been public, several of them urgent. Most were old images from users long gone from the platform, quietly climbing in search over years. BrandViz.AI handed over the list and a design for a weekly scan that catches the next ones automatically.

What changed for NightCafe

NightCafe ended the sprint with a generation system that produces better pages by default, a prioritised map of 578 of them, a measurement layer where there had been none, and a design for turning 10 million monthly generations into a demand signal. What Angus valued most wasn’t any single deliverable.

On how it compared to what they had used before:

Is your organic surface a content problem or a systems problem?

It is a content problem when the pages that would answer your buyers’ questions don’t exist yet, and a systems problem when they exist in their thousands and none of them is good enough to rank. Most AI visibility work is the first kind: find the questions buyers ask, answer them authoritatively, get cited. That is the right approach for most B2B brands, and it is how the majority of BrandViz.AI sprints run.

It is the wrong approach for a platform with 52,000 pages and no capacity to write. There the work is deciding which of the thousand things you could do are the four that matter, then rebuilding the system so every page benefits at once.

At that scale, the question worth asking is whether anyone has checked what the pages you already have are doing.

Common questions about work at this scale