How often runs happen and what they cost
Nothing runs a visibility check on a schedule: every check starts with somebody clicking Run visibility check. This page gives the exact size of a run, whose account pays for it, and every limit that can refuse one. For what to do when a run comes back incomplete, read Why one AI engine returned nothing, or half your prompts were not asked.
When a run happens
- A visibility check runs when you click Run visibility check on the Overview tab, or Run the check with these prompts on the Prompts tab, or Run again on the stale-run banner.
- There is no schedule, no automatic re-check and no background run. A month with no clicks is a month with no new figures.
- One check per workspace at a time. The lock is held by a 75-second lease, so a second tab is told to wait rather than spending your AI account twice.
- The one thing MarqIQ does run daily is a model probe at 06:15 UTC, which asks each provider whether the model AEO measures with is still on its list. Listing models spends no tokens and never touches your key.
Tip
Once the latest run is more than 30 days old, a banner gives its date and age and carries a Run again button. That banner is the only prompt you get; treat a monthly check as the floor for a brand that is actively publishing.
How big one run is
A run asks every active prompt on every connected engine. Nothing is sampled and nothing is rotated, so the size is arithmetic you can do before you click.
calls = active prompts × connected engines
plus 1 call for the sentiment readingForty active prompts on four engines is 160 engine calls, plus the one reading. The run chooser states the engine calls before it starts — "Asks 40 active prompts (3 aided, not scored) on 4 engines — 160 calls." The engines it counts are the ones it can actually ask, so an engine with no usable model is not in that figure.
| What | Value |
|---|---|
| Answer budget per call | Claude and Perplexity are asked for 4,096 tokens, and Groq's gpt-oss models for 6,000 including their reasoning. Every other engine gets its model's own maximum. |
| Time one engine call is given | 20 seconds, or the time the request has left if that is less |
| Time one request spends starting engine calls | 40 seconds, and only with at least 15 seconds left for a call to finish properly |
| Sentiment reading | One call, on a sample of at most 32 excerpts spread across engines |
| Requests per run | As many as it takes — the page sends the next one while calls are still pending, showing "12 of 40 calls done" |
A run that does not fit one request spans several. It is counted against fair use once, when it starts, never again as it continues.
Whose account pays
AEO runs on the AI accounts your workspace connects. MarqIQ does not meter or bill AI work, and does not resell tokens: what a run costs is what your provider charges for those calls.
| Engine | What the provider charges | Key page |
|---|---|---|
| ChatGPT (OpenAI) | Paid per use — a few cents per visibility run. | platform.openai.com/api-keys |
| Gemini (Google) | Free tier included. If your team is in the UK, EEA or Switzerland, Google's terms allow only its paid tier: use a key from a Google Cloud project with billing turned on. Elsewhere, Google uses what is sent to its free tier to improve its products, and AI Enrichment sends contact details to Gemini with this key. | aistudio.google.com/apikey |
| Perplexity | Pro includes credits. The one engine that discloses citations. | www.perplexity.ai/account/api |
| Claude (Anthropic) | Paid per use — a few cents per visibility run. | console.anthropic.com/settings/keys |
| Groq | Free tier included. | console.groq.com/keys |
| OpenRouter | Free models included (rate-limited). | openrouter.ai/settings/keys |
These are the notes and addresses shown beside each engine in Connect AI engines, printed here so the list is usable from a PDF export too.
MarqIQ's included engines
MarqIQ can lend its own free-tier engines — OpenRouter free models, Groq and Gemini — but they are off unless your deployment has them switched on. Where they are off, every AEO call runs on a key you connected, and none of the fair-use limits below can apply to you.
Limits on your own keys
A workspace running wholly on its own keys is spending its own money on its own quota, so MarqIQ does not limit how often it checks. Three limits still apply, and none of them is about cost.
| Limit | Value | What it protects |
|---|---|---|
| Prompts a workspace may hold | 200, active and inactive together | A workspace already above it keeps its prompts and cannot add more. |
| Prompts added per hour | 60 | Counts prompts you type and suggestions you pick. Generate prompts is not counted against it. |
| Checks running at once | 1 per workspace | Held by a 75-second lease, so two tabs cannot double-spend your account. |
Limits on MarqIQ's included engines
These apply only to a request that would spend one of MarqIQ's included keys — that is, one aimed at an engine you have not connected yourself. They exist because those keys are one quota every workspace shares.
| Action | Per workspace, per hour | Per workspace, per day | Shared across every workspace |
|---|---|---|---|
| Visibility checks | 6 | 16 | 336 AI calls an hour, 735 a day, per included engine |
| Prompt drafts (Generate prompts) | 10 | 20 | 12 an hour, 60 a day |
| Recommendation drafts | 10 | 20 | 12 an hour, 60 a day |
| Brief drafts | 30 | 40 | 30 an hour, 100 a day |
| Brand profile analyses | 10 | 20 | 10 an hour, 40 a day |
A day here is the UTC day. The shared budget for checks is counted in AI calls rather than in checks, so a small check spends less of it than a large one.
Take care
A run that would ask more than 20 active prompts is refused on the included engines, before anything is counted. The refusal says how many prompts are active and how many to turn off, and offers the alternative: connect your own AI accounts, where a run can ask every tracked prompt. A run is never cut down to a sample.
Every refusal says when to try again, in the sentence and in the response's Retry-After. A refusal from the shared budget says so plainly, so a wait caused by other workspaces is not read as a penalty on yours.
How long a run stays resumable
- A run that stops part-way stays unfinished, with its answers stored. Resume check continues from the calls still pending and re-asks nothing, so nothing is paid for twice.
- 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 discarded run's answers stay stored but are not shown or scored anywhere. No screen reads an unfinished or discarded run.
If it did not work
A check is refused with a wait, and you are on your own keys
Why: Some engine in the run is one you have not connected, so the request would spend MarqIQ's included key and is counted against the shared budgets.
Fix: Read the message — it names which limit was reached. Connect your own key for that engine, or switch on Use only the AI accounts we own on the AI & Providers settings page so runs never reach for a shared engine.
A run is refused for asking too many prompts
Why: It would ask more than 20 active prompts on MarqIQ's included engines.
Fix: Turn prompts off in the Tracked column on the Prompts tab — the message says how many — or connect your own AI accounts, where the whole tracked set is asked.
Your provider bill is higher than you expected
Why: Every active prompt is asked on every connected engine, so adding one engine multiplies the run by the whole prompt set, not by one call.
Fix: Read the run chooser's first line before each check — it gives the call count. Turn off prompts that are not real buying questions, or disconnect an engine your buyers do not use.
You expected a weekly check and nothing has run
Why: There is no scheduler. Every check is a click.
Fix: Put the check in a calendar reminder alongside your reporting cycle. After 30 days the Overview tab shows a banner with a Run again button, which is the only reminder MarqIQ gives.
Next steps
Related articles
Was this article helpful?