Apify Actor Visibility Monitoring: Apify AI/MCP, SEO, GEO/AEO

AI is becoming the front door to Apify. When someone asks an AI assistant — Claude or Cursor through the Apify MCP server, or Apify's own Apify AI right inside Console — to "scrape Instagram comments" or "monitor my competitors' prices", it searches the Apify Store, reads a handful of Actor cards, and picks one to run. If that pick isn't your Actor, that user was never yours — and you'll never see it happen.

This Actor makes it visible. It runs the same searches an AI agent runs, records where your Actor ranks, simulates which card the AI actually picks, and tells you — honestly — what to fix. Run it regularly: rankings shift, competitors update their cards, and every change you make to your title or description deserves a before/after check.

Now covers Apify AI. Apify's new in-Console assistant, Apify AI, finds and runs Actors from plain-English prompts using the same MCP tools and Store ranking this audit already probes — so your MCP score is your Apify AI visibility, measured the same way. Add SEO and AEO to cover Google organic and the answer engines (Perplexity / ChatGPT / Gemini) as well.

If you…

• published an Actor and wonder why the MCP traffic goes to a competitor,
• just rewrote your title/description and want proof it worked,
• maintain a whole portfolio and want one score per Actor plus a strategy for each,
• are about to build a new Actor and want to know whether the niche is even winnable,

…this is the audit for you.

What you get

• A visibility score (0–100) per Actor, split into the three things that decide AI traffic: does the AI find you, how high do you rank, and does it actually pick your card.
• A breakdown by prompt difficulty — broad ("keep tabs on my competitors"), medium ("monitor competitor websites"), and niche ("track pricing-page changes with Slack digests") — so you see exactly where you're invisible.
• The competitors above you for every query, with their runs, success rate, and pricing — plus the AI's own reasoning for why it picked whoever it picked.
• A strategy per Actor: keep competing, improve the copy, differentiate the card, niche down, pivot, or walk away — an honest market read with concrete next steps.
• Three discovery channels, all in the one flat per-Actor fee (no per-channel charge): MCP / Apify AI (the core — the Apify MCP server and Apify's in-Console Apify AI, which pick Actors from the same Store search), SEO (where your Store page actually ranks on Google page 1 for buyer queries, plus an on-page grade explaining why), and GEO / AEO (do Perplexity / ChatGPT / Gemini / AI Overviews name your Actor when asked your own prompts — with no Apify MCP involved; the full answers are returned).
• One cross-source read when you check more than one channel — the thing no single channel shows on its own, like ranking #1 for agents while no answer engine has ever heard of you.
• An interactive, shareable report page (see a live multi-Actor example) with dark/light mode, filtering and sorting — opened and shared from a link at the end of every run — plus the structured dataset and OUTPUT for spreadsheets or automation, and an optional email summary.

How to run it

• Just add your username (recommended). The run audits every public Actor you own across all three channels — the AI generates realistic test prompts and does the rest. Minimum inputs → full results.
• Or check specific Actors. List one or more Actors (username/name or IDs). If a username is also set, the username audit wins.
• Your own prompts (optional). In "Optional full ownership" the prompts field is a JSON map from an Actor to the exact prompts to test for it — tested verbatim, no AI rewriting. Any Actor not listed there gets AI-generated prompts.

Prompt generation reads only the Actor's public metadata (title, description, categories, README) — the same card an AI agent sees. That keeps the audit reproducible and honest: if the card undersells the use case, the low score IS the finding, and the strategy tells you how to fix the card.

The AI runs on Claude Haiku 4.5 (our included key) by default — zero setup, and its cost is why the fee stays flat. To run Claude Sonnet, GPT or Gemini, paste your own key in "Optional full ownership" and pick the model there. This Actor always runs with a model (it never degrades to a key-free heuristic).

Pricing — flat, per Actor

$0.10 per Actor audited on our included AI key — or $0.05 with your own key (BYOK). Auditing 3 Actors is $0.30 (or $0.15 BYOK). Running more channels never adds a charge — MCP / Apify AI, SEO and GEO / AEO are three lenses on the same audit, all in the one fee. It covers everything: AI prompt generation, every search, the AI pick simulations, the strategy, the shareable page and the email.

Why $0.10 — and why BYOK is cheaper. The fee mainly covers the AI we run for you: Claude Haiku writes the prompts and simulates the picks, and one Claude Sonnet call reads across all three channels. On our included key that AI is on us, so $0.10 is close to cost. Bring your own AI key and you cover that AI — so the fee drops to $0.05. Either way the SEO and GEO/AEO Google checks run on your own Apify account (connect it in "Optional full ownership"), never on ours. An Actor whose audit couldn't complete is never billed.

A fair budget: run it twice a week. Visibility shifts as rivals ship and re-title, so check after each improvement — twice weekly is a good cadence. That's ~$0.80 per Actor per month on our key ($0.40 BYOK) — against $20+/month for a typical SEO rank tracker that doesn't even look at AI-agent or answer-engine discovery. Put the run on an Apify Schedule and add your email in "Email me the results" for a summary with a link to the full report every time.

Advanced: how the measurement works

The two layers of AI selection.

1. Eligibility gate. The Store search the MCP calls returns only FREE / PAY_PER_EVENT / pay-per-result Actors. A rental/monthly Actor is structurally invisible to AI agents — it scores 0 regardless of keywords.
2. Layer 1 — search recall. The agent calls GET /v2/store?search= with short platform/data-type nouns. Ranking is text relevance (name, title, description, README) with popularity as a strong tiebreaker. For broad head terms the top slots are popularity-bound — you can't edit your way past a giant's run count.
3. Layer 2 — the card pick. The agent reads the returned cards (title, description, stats, pricing — not the full README) and chooses one. You can win Layer 2 at ranks 3–8 with a sharper card.

The prompts. The AI generates promptCount (3–8, default 5) prompts spread across broad / medium / niche specificity and varied intents (task, problem, comparison, direct), each with up to 3 semantically different keyword phrasings — synonyms and different noun focus, not truncations. Without an AI key, a deterministic fallback derives prompts from the title and first description clause (tagged heuristic in the data so degraded runs are visible).

The score.

Score = 45% discovery + 25% rank + 30% AI pick

• Discovery: credit per prompt = 70% for being found at all + 30% × the fraction of phrasings that found you (one-exact-phrase visibility is fragile). Broad prompts weigh 1.5×, medium 1×, niche 0.75× — being found from a vague request is the achievement.
• Rank: ranks 1–3 score 1.0 / 0.9 / 0.8 (near-equivalent — the pick decides among the top cards), then quality decays geometrically (rank 5 ≈ 0.45, rank 10 ≈ 0.11).
• AI pick: of the simulated selections that ran, the share your card won. Generated plans simulate one pick per specificity tier; your own typed prompts each get their own.
• When the AI-pick simulation never ran (no key), the score renormalizes over discovery + rank, and the report says the pick was never measured.

A prompt whose Store searches all failed is skipped — an API outage can't masquerade as "not found" — and if every prompt fails, the Actor's audit errors and is never billed. Scores from v0.6 onward are not comparable with earlier versions (the old formula tested near-duplicate keyword ladders).

Roadmap

All three channels are live today: MCP / Apify AI (Store-search discovery + the AI pick), SEO (live Google page-1 rank + a keyless on-page grade that explains it), and GEO / AEO (whether Perplexity / ChatGPT / Gemini / AI Overviews name your Actor for your own prompts), plus the one cross-source read that ties them together. Next: tracking your scores across runs, so you can watch a title or description rewrite move the number.