AI Visibility monitoring is the one part of Agentimus that talks to the outside world. Everything else in the plugin runs quietly on your own site. This feature, when you switch it on and add your own AI keys, actually asks the big AI assistants questions and reports back whether they mention you, link to you, and how you stack up against your rivals — tracked over time so you can see the picture change.

It is built into this plugin — nothing extra to install — but it’s hidden until you ask for it. Because it’s the one feature that spends your own AI credit, it’s an explicit opt-in: turn on Track AI citations under Settings → Features and the AI Visibility tab appears, and a Cited rung joins your AEO/GEO score. Leave it off and neither shows — your score is a clean four-rung ladder.

What it does, in plain terms

You tell Agentimus a few things:

• What you want to watch — your brand, a product, or a person (for example, yourself).
• The questions real people would type — like “what’s the best llms.txt plugin for WordPress?”
• Which AI engines should answer — ChatGPT, Gemini, Claude, and/or Perplexity.

Agentimus then asks each engine each of your questions and reads the answers. For every answer it records three things:

• Mentioned — did the answer name you at all?
• Linked — did the answer point to your website (either in the text or in its cited sources)? This is the strongest signal, because it means an AI is actively sending people to you.
• Rank — when your rivals also show up in the same answer, who gets named first?

It repeats this on a schedule you choose, so you build a history and can watch whether your visibility is climbing or slipping.

Important: it is off by default and uses your own keys

Two things are worth understanding before you turn it on:

• It is the only feature that makes an outbound request. Out of the box Agentimus sends nothing anywhere. This feature is the single exception, and it only ever contacts the AI engines you explicitly enable, only when a check actually runs.
• You bring your own API key (“BYOK”). Agentimus does not include AI access. You sign up with each AI provider you want to use, get an API key from them, and paste it in. Each check spends a small amount of your own credit with that provider. Because it costs your money, Agentimus will never start running automatic checks on your behalf — you have to switch that on yourself.

Your keys are stored on your own server, are used only to run your checks, and are never shown back to the browser unless you deliberately click to reveal one.

Where to find it

Once Track AI citations is on (Settings → Features), an AI Visibility tab appears in the Agentimus menu. Open it. Inside that tab there are two views:

• Results — the scoreboard from your latest check.
• Settings — where you add what to track, turn on engines, paste keys, and set the schedule.

When you first arrive, the Results view will say there are no results yet and point you to Settings. That’s expected — you set things up once, then run your first check.

Step 1 — Add what you want to track

In Settings, the first card is What you’re tracking. Agentimus starts you off with one item, pre-filled with your site’s name and web address so a first check is meaningful straight away. For each thing you track you can fill in:

Field	Required?	What it’s for
What is it called?	Yes	The exact name you want AI to say (for example, Agentimus). Used to detect a mention.
What kind of thing is it?	Optional, but worth filling in	The category a buyer would shop in — “WordPress SEO plugin”, “accountancy firm in Leeds”. This is not what your site writes about; it’s the market you compete in. It’s what the suggested questions (below) are built from.
Its website	Optional	So Agentimus can tell when an AI links to you specifically. Leave blank if it has no site.
Who are its rivals?	Optional	The competing names. Agentimus shows who AI picks instead of you.
What should we ask AI?	Yes to get results	Real questions a person would type. Press Enter after each one. These are the answers that get graded.

The whole screen saves as you type. There is no Save button anywhere on it — your tracked items, your engines and your schedule are all stored the moment you change them. You’ll see a small “Saving…” then “Saved ✓” next to each item, and a plain warning if an item still needs a name or a question before it can be checked.

You can track several different things (up to ten), and each one keeps its own website, rivals, questions, and its own scoreboard. Add more with + Add another. Each item also has an Active / Paused switch — pause one to leave it out of checks without deleting its setup, and turn it back on later.

Rivals and questions are added as chips: type and press Enter to add one, and click a chip to edit it. You can track up to 25 questions and up to 20 rivals per item.

Stuck for questions? Let Agentimus suggest some

Next to the questions field are up to two buttons, and they do different things:

• Suggest questions builds candidates from a set of built-in templates. It’s instant, free, and makes no network request of any kind — no AI is involved. It’s always available.
• ✦ Suggest with AI asks the AI provider you configured in WordPress (under Settings → AI) to write the questions instead. It only appears when a provider is actually set up, and it spends one AI call from your own quota. If the model comes back with nothing usable, Agentimus quietly falls back to the template questions rather than showing you an empty list.

Either way you get a row of suggestions; click one to add it as a question, and ignore the rest. Nothing is added for you.

Both work from the category, and the AI button stays disabled until you fill it in. That’s a deliberate guardrail, not fussiness. A suggestion you accept doesn’t just sit there — it becomes a question that is graded on every single run, forever. When suggestions were built from what a site blogs about rather than the market it sells into, they came out as things like “What is the best JavaScript?” — a question that can never legitimately name you, so it scored a permanent zero and quietly reported that “AI never mentions you”. Better to ask for one short phrase than to hand you a question that is guaranteed to fail.

If you leave the category blank, Agentimus won’t invent a market for you. It offers only questions built from your own name — “What is X?”, “X vs Y” — which are honest, if less revealing.

One thing you’ll notice about the AI’s suggestions: they never contain your own name. That’s on purpose. A question with your name already in it is guaranteed to get your name back in the answer, so it measures nothing.

Step 2 — Turn on AI engines and add your keys

The second card is AI engines. Each row is one assistant. Flip the toggle to turn an engine on, then paste that engine’s API key. You get a key from the provider’s own site — each row has a Get a key link that opens the right page:

Engine	Where you get a key	Answers from
ChatGPT (OpenAI)	platform.openai.com	What it already knows, unless you turn on live web
Gemini (Google)	aistudio.google.com	What it already knows, unless you turn on live web
Claude (Anthropic)	console.anthropic.com	What it already knows, unless you turn on live web
Perplexity	perplexity.ai settings	Always a live web search

Each engine also lets you pick a model. Agentimus defaults to a fast, low-cost model for each provider so recurring checks stay cheap, and you can choose a different one from the list or type a custom model ID.

As with the tracked items, this card saves as you type — flipping an engine on, pasting a key or changing a model is stored straight away.

Test before you rely on it. Each row has a Test button that does one tiny round-trip to confirm the key works. A green “✓ Working” means you’re set; a failure opens a readable message (often with a link to fix it, such as “add billing”).

A saved key shows as dots and is kept safe on your server. Leaving the dots untouched keeps the stored key; clearing the field deliberately removes it. There’s an eye icon to reveal a key if you need to copy or check it.

“But I already gave WordPress an API key”

If you’re on WordPress 7.0 and have already connected an AI provider under Settings → AI — the one that powers Write with AI — it’s fair to ask why AI Visibility makes you paste a key again. The reason is what a visibility check actually grades.

Two things a visibility check needs, and neither is something WordPress’s shared connector can give it:

• The cited sources. A check is graded on which pages each engine cited in its answer — and whether any of them were yours. WordPress’s connector hands back the answer text only; the list of cited sources is dropped before Agentimus can read it.
• Every engine at once. The whole point is to compare — are you mentioned in ChatGPT and Perplexity and Gemini and Claude? The shared connector points at the single provider you configured, not all of them.

Both mean talking to each engine directly, through its own API and its own key. So AI Visibility keeps its keys separate; they stay on your server, encrypted, and are used only to run your checks.

This is intrinsic to the feature, not a WordPress-version thing — even on the newest WordPress, AI Visibility would still connect directly, because the connector can’t return citations. (A useful side effect: because it’s self-contained, AI Visibility also works on WordPress older than 7.0, where the shared connector doesn’t exist at all.)

Step 3 — Live web search (optional but recommended)

By default, most engines answer from their training — that is, what they already “remember” about you. That tells you how AI describes you off the top of its head.

Turning on Live web for an engine makes it actually search the live web before answering and cite the pages it used. This matters because a live answer’s “linked your site” score reflects what AI can find about you right now — not just what it memorised months ago.

• ChatGPT, Gemini, and Claude each have a Live web on/off toggle. (ChatGPT’s live search needs a search-capable model, such as gpt-4.1.)
• Perplexity always searches the live web — there’s nothing to switch on, so it’s simply marked Live web is always on.

Live checks take longer and can cost a little more per question, because the engine runs a real search behind the scenes. Agentimus gives those checks extra time so they don’t cut off early.

Step 4 — Run your first check

Go back to the Results view and click Run check now. Because a full check makes many calls one after another (more so with live web search on), it runs in the background — the button shows “Running…”, you’re free to keep working, and the scoreboard fills in on its own when it finishes. While it runs, your previous complete results stay on screen rather than flickering with half-finished numbers.

If nothing can run yet, Agentimus tells you immediately instead of pretending to work — for example, “Add at least one question in Settings first” or “Enable at least one engine and add its API key in Settings first.”

Reading the results

Once a check completes, the Results view leads with a plain-English sentence such as “AI named Agentimus in 6 of 8 answers, and linked your site in 2 of them.” Below that are four headline cards:

• Seen in answers — the share of answers that named something you track (your overall visibility score).
• Linked your site — the share of answers that pointed to your website. The strongest sign.
• Mentions — how many answers named you, out of how many checks ran.
• Errors — checks that didn’t finish (for example, a bad key or a rate limit). These are shown so a failure never quietly drags your score down.

Under the cards, Visibility over time draws a simple trend line across your recent completed checks, so you can see at a glance whether you’re being mentioned more or less often as time goes on. A couple of checks are needed before the line appears.

Per-item breakdown

Each thing you track then gets its own section showing:

• Seen in answers, Linked its site, and Rank vs rivals for that item.
• Who AI names — you vs rivals — share-of-voice bars comparing how often you were named against each competitor, so you can see who’s winning the conversation.

Question by question

Every question is listed with a small tag per engine so you can see exactly how each assistant answered:

• cited — named you and linked your site.
• mentioned — named you.
• absent — didn’t name you.
• error — the check didn’t complete (the reason is shown right below).

Answers that used a live web search carry a small web marker, and Agentimus lists the actual pages each engine cited as links — so you can see precisely where an AI is getting its information about you. This is often the most useful part: it tells you which of your (or a rival’s) pages the AI is leaning on.

How this feeds the “Cited” rung

When Track AI citations is on, your latest check’s “Seen in answers” rate becomes the Cited rung of your AEO/GEO score — the one rung that measures a real outcome rather than your setup. Two things keep it honest:

• It’s dated, and it expires. The rung notes when it was last checked, and a reading older than 90 days stops counting toward the score (it’s shown as a dated reference until you re-run). An old measurement never quietly stands in for a current one.
• No key, no number. If you remove a provider key — or haven’t run a check yet — Cited reads a grey “—” and is simply left out of the score, its weight shared across the other rungs. Clicking the Cited rung on the Dashboard takes you straight here: to Settings when your setup isn’t complete enough to run a check, or to Results when it is.

Turn on automatic checks (scheduling)

Running by hand is fine, but the real value comes from tracking change over time. In Settings → Schedule:

• Flip Run checks automatically on.
• Choose Daily or Weekly.

These save on their own too — like the rest of the screen, there’s no Save button to press afterwards. From then on, Agentimus runs your checks on that cadence in the background, with no clicks needed. You can still hit Run check now whenever you like.

A few sensible guardrails apply:

• Automatic checks stay off until you turn them on, because each run spends your own AI credit.
• Even when switched on, Agentimus won’t schedule anything until there’s actually something to do — that means the master switch is on, at least one engine is enabled and keyed, and at least one active item has a question. A half-configured setup quietly schedules nothing rather than burning calls on an empty run.
• You can pause a single item or turn off the whole schedule at any time without losing your setup.

Keeping and clearing your history

Under Schedule, Keep history (days) controls how long results are stored before Agentimus trims the oldest ones (anywhere from 7 to 730 days; the default is 180). This applies to every check, manual or automatic.

On the Results view, Clear wipes all stored results for the site if you want a clean slate. Your settings — items, keys, schedule — are untouched; only the past results are removed.

Costs, keys, and privacy

• Costs are yours and are small but real. Each check is one question sent to one engine. If you track 3 questions across 2 engines, that’s 6 calls per run. Live web searches cost a little more than plain answers. Agentimus paces the calls slightly and backs off politely if a provider says you’re going too fast, so a run won’t hammer your account.
• Your keys never leave your server. They’re stored in your own database, sent only to the matching provider, masked when the settings screen loads, and revealed only when you explicitly click the eye icon.
• Results are stored locally, in your own site’s database — nothing is sent to Agentimus or anyone else.
• The providers’ own terms apply to the questions you send them. Their terms and privacy links are listed in the plugin’s readme under External services: OpenAI, Perplexity, Google (Gemini), and Anthropic.
• Removing the plugin cleans up after itself — uninstalling Agentimus deletes the results table, the saved settings (including keys), and cancels the scheduled checks.

Running many sites (multisite)

If you run a WordPress network, there’s a AI Visibility page in the Network Admin menu that rolls every site’s latest numbers into one table — each site’s brand, visibility score, citation rate, engines in use, and last run — with an average across the network. Each site still keeps and configures its own data; the network screen is a read-only overview with an Open link to jump into any site’s own settings.

An honest note on what this can and can’t do

It’s worth being straight about this, because plenty of tools aren’t. This feature measures reality — it does not manufacture it.

• Detection is deliberately simple and transparent: a “mention” means your name or domain literally appears in the answer, and a “link” means your site shows up in the text or the cited sources. It’s an honest floor, not a courtroom verdict, so a result is always explainable.
• Whether an AI spontaneously recommends you for a broad question comes down to authority and reputation, earned over time through genuinely useful content that others reference. No plugin can conjure that, and anything promising “instant AI visibility” is overselling.

Where Agentimus genuinely helps is making sure that when an AI does look at your site, it finds and understands you correctly — that’s what the rest of the plugin (Markdown delivery, structured data, llms.txt, the discovery document, Topics for AI) is for. AI Visibility monitoring is the mirror that shows you whether that work is paying off. Pair it with the Readiness report to fix what’s missing, and the Agent activity log to see which AI crawlers are actually visiting.

Quick reference

Setting	Where	Notes
Show the feature	Settings → Features → Track AI citations	Off by default; the tab + the Cited rung appear when on
Run checks	Enable an engine + add a key	Bring your own key; each check spends your own credit
What to track	Settings → What you’re tracking	Up to 10 items, each with its own name, category, site, rivals, questions
The category	“What kind of thing is it?” per item	The market you sell into, not what you blog about; the suggestions are built from it
Questions per item	Chips in each item	Up to 25
Rivals per item	Chips in each item	Up to 20
Suggest questions	Button under the questions field	Built-in templates: instant, free, no AI call
✦ Suggest with AI	Button under the questions field	Uses the provider from Settings → AI; needs the category; spends one AI call
Engines	Settings → AI engines	ChatGPT, Gemini, Claude, Perplexity
Live web search	Per-engine toggle	Perplexity always live; others optional
Test a key	Test button per engine	One tiny call to confirm it works
Saving	Anywhere on the Settings view	Everything saves as you type — there is no Save button
Run now	Results → Run check now	Runs in the background
Automatic checks	Settings → Schedule	Daily or weekly; off until you switch it on
Keep history	Settings → Schedule	7–730 days (default 180)
Clear results	Results → Clear	Removes stored results only, not your setup