Google was about rankings. AI is about being the answer. This is the whole system that checks if ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews recommend your business, finds out why not, fixes your pages, and earns you into the sources they quote. Every account, every skill, every prompt, step by step.
Read first
This is the one guide in the series with a paid piece: DataForSEO. Signing up gives you free trial credit, and a full live check of 12 prompts across 4 AI engines costs under $2. Everything else is free.
What you're building
One loop that runs your AI visibility
This is part four of the SEO system. Part one (the SEO + GEO Autopilot) fixes your pages for Google, part two (Backlinks) fixes who vouches for you, part three (Content) keeps publishing. This one covers where buyers actually ask now: inside AI chats.
Mirror → Detective → Fixer → Witness → (recheck in 2 to 4 weeks)
And this tracker is different from the rest of the series: the dashboard itself can run real live checks in your browser. You paste one API key, press Run live, and it literally asks each AI engine your buyers' questions and records whether you came up. No coding, no agency tool.
What it costs
Almost free
Free
The tracker file + the seo-geo, seo-dataforseo and seo-content skills
Free
DataForSEO trial credit when you sign up, enough for your first sweeps
<$2
One full live sweep, 12 prompts across 4 AI engines + AI Overviews
After the trial credit you top up pay-as-you-go. We measured it: a real web-searched check costs about 3 cents per engine, and a full 12-prompt sweep across 4 engines lands around $1.50. There is no subscription for what this tracker does.
Step 1 · 10 minutes
Get your DataForSEO key
DataForSEO is the data company behind this whole system. They run a database of 270M+ real prompts people type into AI tools (see the live counter on their site) and they can ask ChatGPT, Claude, Gemini and Perplexity questions through one API.
In the left menu, open API Access (app.dataforseo.com/api-access). You will see two things: your API login (your email) and your API password.
Copy both. Important: the API password is a separate generated password, not the one you log in with. It lives only on that API Access page.
That is the whole account setup. Top up later only when the trial credit runs out.
Step 2 · 5 minutes
Run your first live check
Open the tracker. It loads with demo data so you can see what a filled scorecard looks like.
On the Mirror tab, scroll to the ⚙ Live mode box.
Paste your API login and API password, type your brand name and domain, and press Connect. Your real DataForSEO credit balance appears, that is how you know it worked. (Skipped Connect? No problem, Run live connects by itself as long as the two fields are filled.)
Tick the engines you want, then press Run live ▸ on any prompt row. Watch the chips flip to what the engines really said, and a strip appear under the row listing the exact sources each engine quoted, with a verdict on whether you are one of them. The Detective tab's trusted-sources chart also switches to live data built from your runs.
How the live check actually works, in plain English
When you press Run live, the tracker (from your own browser, your key never goes anywhere except DataForSEO) sends your buyer's question to DataForSEO. They actually ask ChatGPT, Claude, Gemini and Perplexity that question, with web search turned on, and pull a real Google AI Overview for it. The full answers come back and the tracker reads them:
✓ cited means your link was in the answer,
◐ mentioned means your name came up with no link,
✗ quiet means you never appeared.
Every score on the board is computed from those answers, nothing is made up. If a call fails you see an error, never a fake result. One honest gap: Copilot has no public API, so the tracker gives you a copy-paste test question and a direct link to check it by hand.
Step 3 · make it yours
Swap in your business, one paste
The demo tracks a facilities management brand. Give Claude this prompt with the tracker file attached and it rebuilds the prompt list for your business, using real AI search volumes, and keeps live mode intact.
The one-paste master prompt
I'm attaching my AI Search Tracker (search.html). My business: [BRAND], domain [DOMAIN], category [WHAT YOU SELL], customers [WHO BUYS].
1. Using the seo-dataforseo skill (or the DataForSEO MCP), pull the top 12 to 15 prompts people ask AI tools in my category from the AI Keyword Data endpoint, ranked by AI search volume.
2. Rewrite the PROMPTS list in my tracker with those real prompts and volumes, and set my brand and domain as the defaults.
3. Keep everything else exactly as is, especially the live mode, the scoring and the tabs.
4. Give me back the updated file and a one-line summary of the biggest gap you can already see.
Never invent volumes; if you cannot fetch real ones, say so and use the manual route.
In plain English
Wait, what are these skills and pieces?
📊
The tracker (search.html) free
A single HTML file. Double-click it, it opens in your browser and works offline. Your API key is never written into the file, it stays in your browser only, so sharing or publishing the file never shares your key. This is the dashboard you will live in.
The data source. Real prompt volumes from a 270M+ prompt database, plus live answers from ChatGPT, Claude, Gemini and Perplexity through one key. Pay per call, cents at a time.
Teaches Claude the Detective job: can AI crawlers read your site (robots.txt, llms.txt), are your passages quotable, are you named on sites AI trusts. Add it in Settings → Capabilities.
Teaches Claude to speak to the DataForSEO API for you: prompt volumes, live LLM answers, mention lookups. Pair it with the official DataForSEO MCP if you want Claude wired in directly.
The Fixer's toolbox: answer-first rewrites, FAQ schema, the copy structure AI engines actually quote. Same skill the Content Engine runs on.
Step 4
The four moves, and the prompt for each
1 · The Mirror
Shows if AI recommends you. Real buyer prompts, checked live across the engines, scored per engine per prompt.
Prompt
You are my AI search visibility analyst. My business: [BRAND], domain [DOMAIN], category [CATEGORY].
1. Using the seo-dataforseo skill or the DataForSEO MCP, pull the top 15 prompts people ask AI tools in my category, ranked by AI search volume.
2. For each prompt, run it live through ChatGPT, Claude, Gemini and Perplexity with web search on, and check Google AI Overviews via the SERP endpoint.
3. For every answer record: cited (my URL linked), mentioned (named, no link) or quiet, plus every source URL each engine quoted.
4. Output a scorecard: prompt x engine, per-engine visibility %, one overall score. Never invent a result; mark failed calls as failed.
2 · The Detective
Finds why you are invisible: which sources the engines trust, and whether AI can even read your site.
Prompt
Use the seo-geo skill. Take the source URLs my Mirror scorecard collected and audit me against them for [DOMAIN].
1. Rank the domains the engines quoted most, and mark which ones mention me.
2. Check my site: GPTBot, ClaudeBot, PerplexityBot and Google-Extended access in robots.txt, llms.txt, answer-first blocks, FAQ schema, load speed.
3. Give me the missing signals ranked by impact, each with a one-line reason.
3 · The Fixer
Turns every fail into a paste-ready fix: rewritten copy in the length AI prefers, FAQ schema, crawler permissions, llms.txt.
Prompt
You are my AI-readability fixer for [DOMAIN]. For every fail the Detective found, hand me the exact paste-ready fix:
1. A robots.txt block that lets the AI crawlers in.
2. An llms.txt file describing my site in plain language.
3. An answer-first rewrite of my key pages: answer in the first 25 words, one concrete number, no marketing fluff.
4. FAQPage schema with my real customer questions.
Tell me exactly where each snippet goes, and keep my voice.
4 · The Witnessthe long game
You cannot edit what ChatGPT says, so you earn your way into what it reads: the Reddit threads, directories and best-of lists the engines already quote.
Prompt
You are my citation-earning strategist. From the sources my scorecard collected:
1. List every source I am not on, ranked by citation count x realistic effort (Reddit threads, G2/Capterra style directories, trade press, best-of listicles).
2. For the top 5, draft the exact contribution in my voice: a helpful Reddit reply (value first, never spam), directory blurbs, a 2-line press pitch, an author outreach email.
3. Rules: never astroturf, disclose affiliation where required, and remind me engines re-crawl slowly, recheck in 2 to 4 weeks.
Honest note: this is the slowest move and the strongest one. Fixes show up in days, placements take weeks to get picked up. The tracker's recheck cycle is how you see it land.
AI answers shift constantly. Schedule one small task in Claude (Settings → Scheduled Tasks) so the recheck happens without you.
Prompt
Every Monday at 8am, re-run my AI visibility check: my 12 tracked prompts across ChatGPT, Claude, Gemini and Perplexity via DataForSEO, plus Google AI Overviews. Compare to last week's scorecard. Message me only if my visibility score moved, I gained or lost a citation, or a new source started appearing in the answers.
If you would rather have this running on your real business by the end of a call, prompts tuned to your buyers, tracker rebuilt for your brand, first live sweep done together, book a paid working session with me.