15,500+ descriptions and metas generated, from 20 manual minutes to a 90-second run
- Client
- A multi-brand casino operator
- Sector
- iGaming
- Capabilities
- Automation & AI
- Stack
- Airtable · n8n · LiteLLM · OpenAI · Anthropic · Google Gemini · Casino platform APIs
3 casino brands, 1,150+ games, 3 languages: every game description was written and translated by hand. We built a pipeline the marketing team runs themselves.
The operator runs 3 online casino brands on a shared catalogue of 1,150+ games. Each game can carry up to 3 brand variants (about 2.3 on average, since not every game is published on all 3 casinos), and each variant needs a long-form description and a meta description in 3 languages.
At the time of the project, all of that content was produced and maintained by hand: written per game, per brand, per language, translated in a separate manual workstream, then pushed into each casino platform brand by brand. Each brand has its own voice, from how it addresses players to its slogan and positioning, and that voice was carried by the people who wrote it, not yet captured in a system.
The catalogue grows by about 15 new games per month.
Multiplied across casinos and languages, the catalogue means 15,500+ pieces of content to produce and keep current: 7,800+ long-form descriptions and as many meta descriptions, more than any copywriting team could realistically manage. At 15 new games a month, a manual process means a permanent backlog and copy that drifts out of date the moment it ships.
And the problem compounds: the catalogue is projected to pass 2,000 games within 5 years, which at the same coverage average means 4,500+ game descriptions to maintain and 27,000+ generated pieces across languages and formats. The operator did not need a copywriting sprint; it needed an engine built for that scale.
There was no quality safety net: no automated checks on word counts, meta lengths, banned words, required headings, or game-name usage. Brand voice and the regulated wording online gambling requires were enforced by manual proofreading, costly at that volume and impossible to keep consistent at scale.
Publication was slow and hard to verify: with no central record of what had been published, when, or in which language, gaps such as missing locales, stale content, and partial pushes could surface only after content was live on the casino site.
We built a 3-layer system: an Airtable workspace holding the data, the configuration, and the review interface, an n8n orchestration layer running the workflows, and a LiteLLM gateway routing generation and translation across OpenAI, Anthropic, and Google models. The full catalogue of 1,150+ games and 2,600+ content records was migrated in and UUID-linked at the start.
Each brand’s voice is configuration, not code: 31 parameters per brand covering tone of voice, slogan, brand DNA, allowed analogies, allowed calls to action, forbidden words, and length ranges. Marketing edits them directly in Airtable and the next generation run picks them up. Anti-repetition rules (banned openers, banned filler formulas, randomized opening styles) keep the same game reading differently on each of the 3 brands.
Quality control is built into the pipeline: 7 scored checks on every generation, covering word count, meta length, forbidden words with zero tolerance, required headings, game-name usage, and JSON integrity, backed by a 3-attempt retry loop that feeds failures back as prioritized corrections. Every record carries a quality score and a full scorecard, and after tuning, validation success on the test sample went from 50% to 100%.
The marketing team runs the whole machine from 3 Airtable buttons: generate, regenerate, publish. Content is generated in Markdown, where structure stays checkable, then converted to clean HTML at publication, so each casino platform’s API receives exactly the markup its pages expect. Once the team approves a piece, one click triggers the publication workflow and pushes it to the live site in all 3 languages, with live progress pages and readable error messages. A health dashboard completes the loop, giving the team a live view of database health: coverage, gaps, and the state of every record at a glance.
The entire game catalogue now lives in Airtable, and the marketing team generates every description itself, against the up-to-date tone of voice of each casino universe: 7,800+ long-form descriptions and as many meta descriptions across the 3 languages so far, 15,500+ pieces of content, about 13.6 generations per game.
By hand, even with an AI assistant in a chat window, one game on one casino is about 20 minutes of work with errors slipping through; the pipeline delivers it in about 90 seconds with the 7 checks built in, and the human contribution shrinks to a 30-second review, about 10 times faster.
A game across all 3 brands takes about 3 to 4 minutes, 10 games run unattended in about 30 to 45 minutes, and platform cost is fixed at about EUR 48 per month. Every new game flows through with no extra setup and no developer in the loop.
See what one system would change
A free 30-minute call: we walk through where your revenue leaks and what fixing it actually takes. No pitch.
