See whether AI assistants recommend you
A visibility check asks the questions your buyers type into AI assistants, on the AI accounts you connect, and records whether the answers name you. This article takes you from an empty screen to a finished first run. To read the numbers that run produces, go on to How to read your AI visibility score.
Before you start
Plan required
AEO is on the Business and Enterprise plans. Below those, the AEO pages show an upgrade panel with Try it free for 7 days. One click, no card, and one trial per feature per workspace, ever — AEO and AEO Advanced are counted separately, so trialling one leaves the other available.
Permissions required
Connecting and disconnecting AI engines is admin-only, and a read-only workspace is refused. Everything else on these pages needs no extra rights.
Note
AEO runs on the AI keys you connect. MarqIQ can also lend its own free-tier engines, but they are off unless your deployment has them switched on. Connect at least one key of your own before you start, and allow a few minutes for the provider sign-up.
Connect an AI engine
Nothing can be measured until one engine is connected. Run visibility check stays greyed out until then.
Click AI Answer Visibility in the left-hand menu.
The first time you land here in this browser with nothing of your own connected, Connect your trusted AI opens on its own. Afterwards you open it from the button.
Click Connect AI engines.
The count beside the button is the engines that can be asked right now. Green means at least one. Red means none.
Click Open key page beside the engine you want.
Six are listed, in the order in the table below. Each row says how to log in and what the provider charges.
Paste the key into the field beside that engine.
The field names the provider's own product: Paste your OpenAI platform key, Paste your Google AI Studio key, and so on.
Click Verify & connect.
MarqIQ checks the key live against the provider's model list, which spends no tokens, and refuses a key that cannot reach the model AEO measures with.
Click Done — continue.
Keys are stored encrypted and never shown back. To remove one, reopen the modal and click Disconnect.
| Engine | Where to get the key | What it costs |
|---|---|---|
| ChatGPT (OpenAI) | platform.openai.com/api-keys | Paid per use — a few cents per visibility run. |
| Gemini (Google) | aistudio.google.com/apikey | Free tier included. If your team is in the UK, EEA or Switzerland, Google's terms allow only its paid tier, so use a key from a Google Cloud project with billing turned on. |
| Perplexity | www.perplexity.ai/account/api | Pro plans include monthly API credits. The one engine that discloses citations. |
| Claude (Anthropic) | console.anthropic.com/settings/keys | Paid per use — a few cents per visibility run. |
| Groq | console.groq.com/keys | Free tier included. |
| OpenRouter | openrouter.ai/settings/keys | Free models included, rate-limited. |
The same six addresses are behind the Open key page links in the modal. They are printed here so the list is usable from a PDF export too.
Three keys are checked differently
OpenRouter's model is picked fresh on every run, so its key is only checked as a working key and not against a model. Perplexity is proved against an authenticated read instead, because its model list answers to any key at all. And a model list that answers but cannot be read lets the key through. In those three cases a key can be accepted and still fail on a run, so read the first run's health line before you trust it.
Tip
Connect more than one engine. A brand that Perplexity names and ChatGPT never does needs a different content plan from one that neither names, and only a per-engine number tells you which you have.
Citations needs Perplexity
Perplexity is the only engine that discloses the sources behind its answer, so the Citations tab needs that key specifically. Citations and Recommendations are part of AEO Advanced, which is on Enterprise — or a 7-day trial you can start from the upgrade panel on either page.
Set up your brand profile
The profile is what every check looks for: your name in the answers, and a plain-English summary of what you sell that the question drafting works from.
Type your website into Website domain.
The hostname alone, such as yourcompany.com. A domain carrying a port, a login, a backslash or a space is refused before the site is read.
Type your brand name into Brand name.
Leave it blank to take the name from the site's title. A typed name is capped at 120 characters. A longer one is refused rather than cut, because every answer is matched against it.
Type your rivals into Competitors, separated by commas.
Up to eight. You can leave this blank and add them later on the Brand profile card.
Click Analyse website & set up.
MarqIQ reads the title, description, headings and social profile links, then asks an AI to write two or three sentences on what you offer and to whom. The result is stored and shown on the Brand profile card.
Take care
Connect an engine before you click Analyse website & set up. The setup card appears only while the workspace has no profile, and there is no re-analyse button afterwards. With no engine reachable the drafting fails, the profile still saves, and the Brand profile card reads "No offerings summary yet." with no way to redraft it on screen.
Choose the questions
A check needs at least one active prompt. A drafted set of ten is enough for a baseline.
Click Prompts.
Click Generate prompts.
MarqIQ drafts up to ten category-level buyer questions from your offerings summary. None of them names your brand, and any draft that does is dropped before it is saved.
Turn off anything you do not want asked, using the switch in the Tracked column.
Tip
Choose the questions to track covers what makes a question worth measuring, how to add your own, the caps, and why a question naming your own brand is labelled aided and left out of the score.
Run the first check
Click Overview.
Click Run visibility check.
Read the panel headed "Before this run: which prompts should it ask?"
It states exactly what the run will ask — active prompts, how many are aided, how many engines, and the total calls — and whether that set matches the last run's.
Click Keep the same prompts and run.
On a first run there is nothing to compare with, and the panel says so. When the set already differs from the last run's, the button instead reads Run with the current prompts (not compared with the last run).
Leave the tab open while the bar fills.
The count reads "12 of 40 calls done", then "all 40 calls done — reading sentiment and scoring the run". A run that does not fit one 60-second request spans several, so a wide prompt set takes a few minutes.
Note
Every active prompt is asked on every connected engine. There is no sampling and no rotation. Movement between two runs is therefore a real change, not a different sample.
How the questions are asked
Prompts go to the engines verbatim, with no system framing and no hint that your brand exists. Whether you were mentioned is decided from the answer text, never asserted by an AI grading its own answer. How MarqIQ decides you were mentioned sets out both rules.
The run is finished when the tiles fill in. How to read your AI visibility score says what each of them counts.
If it did not work
Run visibility check is greyed out
Why: One of four things: no engine can be asked, a change is still saving, a check is already running, or an unfinished check is too old to resume.
Fix: Hover the button. The tooltip says which — "Connect an AI engine first.", "No connected engine can be asked — see the note above.", "Wait for the current change to finish saving.", "A visibility check is in progress." or "An unfinished check is too old to resume — discard it below to start a new one."
Clicking Run visibility check answers "No AI engines are connected yet"
Why: Every key has been removed, or the workspace has Use only the AI accounts we own switched on with nothing connected under it.
Fix: Click Connect AI engines and connect one key. Check the AI settings page if the switch is the cause — the last own key cannot be disconnected while that setting is on, so this usually means the setting was on before the key went.
A key was accepted at connect time and fails on every run
Why: That engine is one of the three MarqIQ cannot check against a model — OpenRouter, Perplexity, or any provider whose model list answered but could not be read. The key itself was proved. The model behind it was not.
Fix: Read the amber banner above the tiles — it names the engine and the provider's own reason. Connect a second engine so the next check still has something to measure with. Why one AI engine returned nothing, or half your prompts were not asked covers each reason.
Setup refuses your domain
Why: The field takes a hostname only. A port, a login, a backslash or a space is refused before the site is read, so the attempt costs nothing.
Fix: Type the domain on its own — yourcompany.com — with no scheme, path, port or credentials. The message reads "Enter the website domain only, without a port or login, e.g. yourcompany.com".
The profile saved but the Brand profile card says "No offerings summary yet."
Why: No AI engine was reachable when you clicked Analyse website & set up, so the summary could not be drafted. The profile still saves, so your domain and competitors are not lost.
Fix: Connect an engine, then ask support to redraft the summary. There is no button for this on the page and no way to clear the profile yourself. Write prompts by hand on the Prompts tab in the meantime — only the drafting needs the summary, not the run.
The tab closed mid-run and the numbers never appeared
Why: A run that stops part-way stays unfinished. Its answers are stored but no screen quotes them, because every reader takes finished runs only.
Fix: Reopen Overview and click Resume check beside the run button, or Resume on the amber banner. It continues from the calls still pending and re-asks nothing. After 24 hours it can no longer be resumed — its answers would span two days and be scored as one measurement — and the only option is Discard.
A second tab says a check is already running
Why: One check per workspace at a time, held by a 75-second lease so two tabs cannot double-spend your AI account.
Fix: Wait for the time the message gives. While another request holds the run, the page re-checks every 10 seconds and Discard reads "Being worked on".
Next steps
Related articles
Was this article helpful?