Shipping TechPulse: a keyword-research pipeline play-by-play

How I went from “iPhone 18 sounds like an SEO opportunity” to a live editorial Apple site — the data, the decisions and the parts I’d do differently.

TechPulse (techpulse.press) is the most recent project I’ve shipped, and the first one where the upstream research was more interesting than the site itself. So this post is mostly about that pipeline and what it took to turn it into a live site without making any of the obvious mistakes.

Step one was deciding whether the opportunity was real. I ran the iPhone 18 keyword cluster through my standard scoring: monthly US volume, average keyword difficulty, intent mix, brand / trademark risk, and seasonality profile. The numbers came back unusually clean: 100 keywords, 256K monthly volume, average KD 19.1, 81% informational intent, and a strong seasonal peak every September that maps onto Apple’s launch window.

The brand had to be Apple-safe. No “iPhone” or “Apple” in the domain (trademark risk), no model-year anchor (iPhone 18 becomes old in 24 months), short and pronounceable, and on a TLD Google already trusts for news / editorial. After a few hours of brainstorming and one false start, I landed on techpulse + the .press TLD. Two syllables, no trademark conflict, future-proof for iPhone 19, iPhone Fold and visionOS coverage.

Content architecture was decided by intent. News and rumor pages for time-sensitive stories with a status field (Confirmed / Credible / Possible / Speculative) and a primary-source link on every claim. Spec hub pages for each device with a structured spec table. FAQ pages for question-form queries that target “People also ask” and featured snippets. Compare / Best pages for commercial-intent queries. The first sprint shipped 36 pages covering iPhone 18 lineup, iPhone Fold and iOS 27.

E-E-A-T was a hard gate, not a checklist. Every article lives in MDX with a frontmatter status machine (draft → review → approved) — only “approved” can ship. Changes go through Git pull requests; the diff is the audit trail and the editor’s approval is a literal review on the PR. AI assists with drafting and editing only. No AI-generated quotes, no AI-invented data, no auto-publish.

For tech, I went static-first: Next.js (App Router), Tailwind CSS, Cloudflare Pages for global delivery, Pagefind for client-side search, Beehiiv for the newsletter, Plausible for privacy-respecting analytics. The same MDX source feeds the site, the RSS feed and any future partner syndication.

What I’d do differently next time: build the site 4 months ahead instead of 3 (the iPhone 18 cluster has more headroom than I realised), and start the Cancel Tracker from day one — once a city starts debating an ALPR exit, the conversation moves fast.

If you’re thinking about something similar, the single biggest unlock for me was treating keyword research as an upstream product, not a one-off SEO task. It changes what you ship, what you name it, what you build on day one, and what you don’t bother with at all.