Ask the page. Ship the fix
The AI panel works the live page in your tab: your render, your Search Console, the real SERP. Then it hands back the exact change to ship.
From questionto fix
Everything below happens in the tab you already have open. The panel reads that tab’s render, not an export of it.
Ask
“Can you spot any issues with this page?” The panel reads the render in your tab, not a cached crawl, and pulls the queries the page already earns from Search Console.
The diagnosis
Two flags: a title written for the brand instead of the “linen duvet cover” queries your clicks come from, and a canonical pointing at ?page=1. One is a rewrite, one is a template bug.
Your call
It proposes the rewrite, then asks before it ships anything: update the title in Shopify, file the canonical with your devs in Linear.
Shipped
One “go ahead”. The title goes live through the Shopify connection and the canonical lands in Linear as AUR-212 with the failing rule attached. The title rule re-runs on the next load.
Make AI answersauditable
Every claim in an answer traces to the tool result that produced it: the Search Console row, the audit rule, the SERP position. What a chat tab asks you to trust, the panel lets you check.
Your data first
It starts with your own numbers: Search Console shows France down 18%, and an audit of your /fr/ pages flags hreflang missing.
Stand where your buyers stand
It moves the browser to a Paris exit node and reads the results page your buyers get. You’re 5th now, not 3rd.
Read the competition
It opens La Redoute’s page and runs the same checks. Their French hreflang is complete; yours is missing.
Check every claim
Four sources, one answer, and every sentence cites the receipt it came from. Fix the template and it rechecks the 12 flagged pages.
| A chat tab | The panel | |
|---|---|---|
| What it reads | Training data, and whatever you paste in | Your rendered page, your Search Console, the live SERP |
| Where it stands | Nowhere | 50+ cities, through real exit nodes |
| How you check it | You proofread it | Every claim cites the tool result behind it |
| What you get | Advice to carry somewhere else | The fix, applied and re-scored in the same tab |
The panel,up close
The skills menu, the prompt in front of every action that changes a page, keyword data on a competitor, a title rewrite against its pixel limit, and the SERP from another city.
16 ways of working. Type / and pick a skill: a page audit, a SERP read, a keyword gap against a competitor, your visibility in AI search. Each is a method the panel follows, with the tools it needs already wired in.
It asks before it acts. Anything that changes a page or the browser waits for your go-ahead, and one click halts a run mid-tool. The transcript keeps everything it did up to that point.
Keyword and competitor data, built in. Ask who you compete with and it finds them, compares the keywords you share, and pulls volume and difficulty for the gaps. Lookups use your team’s AI credits.
Ask for the fix and it writes it. A title inside its pixel limit, schema validated against the rich result it targets, a whole page rewritten against the render. The same tools the panels use, pointed at the change.
It reads the live SERP. It opens the live results page you rank on and reads it like you would. Ask for another market and it moves the browser’s location itself.
Also
16 skills
Any open tab with @
Permission per action
AI credits included
Your other tools,in the same conversation
The AI panel is an MCP client. Connect Ahrefs, GitHub, Linear, or any server that speaks the protocol, and its tools sit next to ours. Bring over the connections you set up in Claude or Codex, or ask the panel to add one from chat.
An AI with the page open.And the tools to change it.