Agent V and Support
The assistant panel in the dashboard holds two agents. V works on your visibility: it reads your audits and your site, drafts fixes, and proposes changes to your account for you to confirm. Help answers questions about the product and your account, and passes you to a person. Both run on Anthropic's Claude models and can only see the account you are signed in to.
Opening it
- Ask V, the floating button, or Ctrl/⌘ + J, opens the V tab.
- Help icon → Ask support opens the Help tab.
- The panel appears once the account has a brand, and works on the active brand. "Looking at" shows the screen and brand it was opened from.
- An open conversation is restored for 72 hours. Start a new conversation resets it.
What V can do
| Capability | Tools | Needs your confirm |
|---|---|---|
| Read your plan, usage, brands, audits, scores, recommendations, accuracy flags and keyword results; compare two audits | account.overview, account.list_brands, audit.get_latest, audit.list, audit.compare, audit.list_recommendations, audit.list_accuracy_flags, audit.list_prompts | No |
| Read a page on your own site as text | site.read_page | No |
| Open a dashboard page beside the conversation | dashboard.open | No |
| Draft page copy, robots.txt lines or JSON-LD to paste | fixes.draft_copy, fixes.draft_robots, fixes.draft_schema | No |
| Run an audit on every active keyword and every platform your plan allows (uses one from your allowance) | checks.run_audit | Yes |
| Check one page on your site | checks.check_page | Yes |
| Add keywords, add a competitor, mark a recommendation done | write.add_keywords, write.add_competitor, write.mark_recommendation_done | Yes |
A change appears as a card, "V wants to", with Confirm and Not now. Nothing changes until you confirm. Drafts are text for you to paste; V never edits your site. Drafted JSON-LD is for keeping markup in agreement with the page; structured data is not an AI ranking lever.
V's starting points: Explain my score, Fix one recommendation, What are they getting wrong, and Ask anything.
What Help can do
Help searches this documentation and the help center, reads your plan and latest audits, and files a support ticket. Talk to a person under either tab files the ticket directly with the conversation so far, with no AI involved. The panel confirms with the ticket number and the address we reply to, and the button changes to "A person has this".
Limits
V conversations are counted per calendar month (UTC). One conversation counts once, however many messages it has. The panel shows "N of M this month".
- Free — 5 V conversations a month, each capped at 60,000 tokens
- Starter ($49/mo) — 50 V conversations a month, each capped at 120,000 tokens
- Growth ($99/mo) — 200 V conversations a month, each capped at 200,000 tokens
- Pro ($149/mo) — 600 V conversations a month, each capped at 300,000 tokens
- Help conversations do not use that allowance. They are limited to 10 new conversations a day and 30 messages an hour.
- A message is up to 4,000 characters.
- Talk to a person can be pressed 5 times an hour; pressing it again returns the open ticket.
- Starting V conversations is also rate-limited per hour.
Errors
| Message | Meaning |
|---|---|
| "You have used all N assistant runs on the [plan] plan this month." | V allowance used (HTTP 403, QUOTA_AGENT_RUN_LIMIT). Resets on the 1st, UTC. |
| "AI checks are paused: the platform's daily AI spend limit has been reached…" | A platform-wide spend limit (HTTP 503). Does not use your allowance. |
| "You have started 10 support conversations today…" | Help's daily limit. Continue an open conversation instead. |
| "Could not reach a person. Email hello@99visibility.com…" | The ticket could not be filed. The email reaches the same team. |
Data and isolation
Each tool runs with your own signed-in database permissions, pinned to one brand, so the assistant cannot read another account. site.read_page and checks.check_page fetch only your brand's own domain, and page text is treated as data, not instructions. Your messages and what the tools read (account, audit and page data) are sent to Anthropic to produce answers.
Need a person rather than the assistant? Every help center article ends with how to reach one.