How TriRank writes content and prepares it for publishing

What actually runs when an article is generated

The content steps below, in order, are what autopilot runs. Generating an article yourself runs the first two and stops there — no internal links, no cover image, nothing sent to WordPress. Both paths check your credit balance and bill the run the same way; autopilot also picks the target keyword, which you type in yourself when you generate an article by hand.

1. 01
   WriteClaude writes the draft against your target keyword, in the niche and brand voice you set. If you set neither, it falls back to a generic voice.
2. 02
   Quality gateOne gate, applied to every draft — the main checks are detailed below. A draft is never edited into shape to get past it. Under autopilot a failed draft is regenerated from scratch; when you generate an article yourself it is not regenerated — the failure is reported to you and no credits are spent.
3. 03
   Internal linksLinks into your own published articles, matched on whole words, never inside headings, code blocks or existing links. Each article is linked at most once. If a phrase does not already occur in the text, no link is forced.
4. 04
   Cover imageAn original illustration, generated per article. No stock photos. If generation fails, a plain branded placeholder is used instead.
5. 05
   WordPress draftThe article is sent to your WordPress site as a draft. You review it and press publish. We do not publish for you.

The quality & ready-to-publish bar

Nothing ships that we wouldn't put our own name on. Every draft passes an automated quality gate — minimum length, clean structure, and keyword coverage — and is written in your brand voice. A draft that fails the gate doesn't ship.

How results are measured

We connect directly to Google Search Console and read your real numbers — clicks, impressions, and rankings for your site. AI citations are tracked by monitoring whether assistants reference your site for relevant queries. We never fabricate or estimate these figures — if it isn't in your own analytics, we don't claim it.

The AI Visibility Score (0–100)

This weighted 0–100 score powers the TriRank badge and the “actually cited” layer of your free-audit report — it measures how often AI engines cite you, weighted by each engine's reach. (Your report's structure score is a separate number, grading whether your site is built to be cited.) A weighted score is only as honest as its inputs, so here is exactly how it is computed — every input, its source, and its limits.

Why these four engines

We probe ChatGPT, Claude, Gemini (Google AI) and Perplexity. Measured by referral-traffic share, these four together account for the overwhelming majority of measurable AI referrals. Many products marketed as separate "AI tools" are wrappers built on these same underlying models — they inherit the underlying citation behavior, so measuring them separately would double-count.

The weights

When an engine cites your site, its verdict contributes its weight to the score:

Source: Goodie, “2026 AI Search Traffic Report” — B2B AI referral share, brand-averaged, March–April 2026. The original values are 62.6 / 18.5 / 10.6 / 7.3 (which sum to 99); we normalize them to 100.

Limitations

How we probe

• Each domain is tested with one representative brand prompt, run across all four engines in parallel.
• A citation verdict is based only on the structured source URLs the engine returns — never on the wording of the answer text.
• If an engine returns no verdict, that engine is left out of that check rather than counted either way.
• A single probe is a snapshot. AI answers fluctuate — that is why the badge carries the date it was checked.

What this score is not

It is not a traffic prediction, not a ranking, and not a measure of your visibility across the entire AI web. It measures one thing: on one brand query, which of the four major AI engines cited your site — weighted by each engine's share of AI referral traffic.

Methodology FAQ