The score on our AI search visibility checker climbs when you tick boxes, and that is the whole mechanism. It never queries ChatGPT, never counts citations, and never sees your traffic. It is a self audit with fixed arithmetic, and it helps only when you know what the 16 checks weigh and what they ignore.
The 16 checks and the arithmetic behind the score
The checklist holds exactly 16 items in four equal categories. Content covers citation worthy material, clear answers, FAQ sections with schema markup, and author expertise signals. Structure covers JSON-LD, semantic HTML, heading order, and tables and lists that parse cleanly. Technical covers page speed, accessibility, robots rules that admit AI crawlers, and a submitted sitemap. Authority covers backlinks, social signals, brand mentions, and consistent name, address, and phone data.
The score is one line.
const score = total === 0 ? 0 : Math.round((checkedCount / total) * 100);
Each check is worth 6.25 points. Four checks land on 25 percent, eight on 50 percent, twelve on 75 percent, and sixteen on 100. The gauge colors follow fixed thresholds, green at 75 and above, amber from 50, orange from 25, rose below. The progress ring draws from a radius of 52, with the stroke offset computed from the circumference, so the arc length tracks the percentage directly.
Your ticks persist in browser storage under the key ai-visibility-checks, and the reset link clears the record in one click.
The prompt tester and its one trap
The side panel builds six prompt templates, including best-of questions, how-to questions, a comparison, an expert trust question, and one template that hardcodes the year 2026. You type a topic and an optional site URL. Each template receives the topic, and the URL is appended inside a bracketed suffix so the engine sees which site you care about.
The trap sits in the substitution line. An empty query field falls back to the interface placeholder, the literal string e.g. project management tools. Skip the input, copy a prompt, and you ask an AI engine about a string that starts with e.g. Fill the topic field every time, or your test session measures noise.
Copy uses the clipboard API with a fallback path that creates an offscreen textarea and runs the legacy copy command, and the copied label shows for 1500 ms. The panel links out to four services, ChatGPT, Perplexity, Google Gemini, and Microsoft Copilot, so one prompt can run in all four without retyping.
How to run AI search visibility checks that mean something
The tool prepares prompts. The method is yours, and this sequence produces comparable results.
1. Fix one topic query and keep the wording identical across every run.
2. Include your brand name in a second variant, since attribution is a harder signal than bare presence.
3. Run the same prompt in all four linked services and record which domains each answer cites.
4. Log your own domain as cited, mentioned without a link, or absent.
5. Repeat weekly on the same weekday and compare against the previous log, not against your impression.
6. Tick checklist items only when the underlying change is live on the site, never for planned work.
One prompt across four engines beats four prompts across one engine, because the engine differences are the variable you want to see.
What the score cannot tell you
State the limits plainly. The score counts self reported boxes, so a site can read 100 percent while no engine cites it. AI answers vary between runs, so a citation this week can vanish next week with no change on your side. The checklist contains no measurement, no crawler, no timing test, and no backlink count, so the technical boxes still need real audits behind them. Progress lives in one browser, which means the score does not follow you between machines, and a shared computer exposes your audit state to the next user. The year 2026 hardcoded in one template will also age, so rewrite that prompt manually once it stops matching your intent.
Checklist before you trust a high score
- [ ] Confirm every ticked item is live on the site, not scheduled.
- [ ] Fill the topic field, and never test with the placeholder string.
- [ ] Log citations from all four services weekly with fixed wording.
- [ ] Track brand name attribution separately from plain mentions.
- [ ] Re-run the audit after each structural change, not on the calendar alone.
Sixteen boxes, four engines, and a weekly log beat any single number. Open the [AI search visibility checker](/en/ai-search-visibility), run the four engine pass on your own topic, and send us the citation counts. If your score and your citations disagree in an interesting way, that gap is exactly the data we want to see.
