How we write
Every article on EAC follows the same four-step process. We publish this methodology because trust on the open web is now scarce, and readers deserve to know how the answer they are reading was produced.
1. Research
Before drafting, an editor compiles a research brief: the precise question to answer, the audience (visitor / resident / professional), the authoritative sources, and the recent regulatory or commercial changes worth noting. For a fare-related article that means TfL’s current fare table; for a Knowledge piece, the latest syllabus revision; for tipping, multiple cabbie surveys cross-checked against tourist reports.
2. Drafting
We use large language models (currently Google Gemma 3 27B, run locally on our own hardware) to produce a first draft from the research brief. The model is instructed to:
- Cite specific facts (years, fares, place names) rather than vague generalities.
- Avoid the small set of phrases that mark machine-generated text (“in today’s world”, “moreover”, “delve into”, “plays a crucial role”).
- Vary sentence and paragraph length.
- Use British English spelling and turn of phrase.
A draft this stage is not publishable. It is raw material for editing.
3. Editing and fact-checking
An editor reads the draft against the research brief and the source material. Every concrete claim — a fare in pounds, a year, a piece of regulation, a place name — is checked against the cited source. Errors are corrected. Padding is cut. Sentences that sound machine-written are rewritten. The article is structured so the reader can scan, with clear H2/H3 headings and a closing FAQ section.
4. Publication and maintenance
Each published article carries:
- A “Last updated” date at the foot.
- A list of sources used.
- Structured data (JSON-LD
Articleand, where appropriate,FAQPageandBreadcrumbList) so the page is machine-readable.
When a source changes — TfL revises a fare, gov.uk updates a regulation, the Knowledge syllabus is amended — affected articles are revised within about two weeks. The “Last updated” date and the changelog at the foot of the article record the revision.
What we will not do
- Publish AI output without human editing.
- Invent author personas with fake biographies.
- Use AI-generated photographs of named real-world products (a specific TX5, a particular driver) — we use Ernie Image-Turbo for atmosphere shots only.
- Quote anonymous internet posts as fact.
- Run affiliate links to booking platforms.
- Take advertising before the site has a stable, organic audience.
If any of those rules feel restrictive — they are, by design.
Last updated 10 May 2026.