TL;DR
- Local Mode AI visibility checks now run in parallel lanes, one per AI engine, instead of one at a time, so a full run finishes in a fraction of the old wall-clock time.
- Google AI Mode and Google AI Overviews are checked at the same time now, and the AI Overviews check stops waiting early when a result plainly has no AI Overview in it.
- The entity graph detects two new kinds of relationships: entities that keep showing up on the same pages together, and entities whose meanings sit close to each other. Both are computed on your machine or optionally using a paid frontier model.
- Send Moose can now pull live data from the Google Search Console API mid-job, and he reads a site's llms.txt file and MCP manifest for extra context.
- Paid managed plans got more usage at the same price: model multipliers were re-anchored to current list prices, and most models now draw 25 to 70 percent less from your monthly allowance, several far less.
- Two smaller fixes: scheduled visibility runs that only use local-fetch engines now send their summary email, and paid plans can remove the attribution line from the audio embed.
Every engine gets its own lane
Local Mode visibility checks ask the AI engines about your brand from your own machine, with a real browser, prompt by prompt. This week the run got lanes: ChatGPT, Perplexity, Grok, and both Google checks now each work through their own queue, side by side. Inside a lane there is still a short, courteous pause between checks, because nobody likes being hammered. Across lanes, everything moves at once, and the full run wraps in a fraction of the wall-clock time.
Two Googles at once
Google gets checked two ways: AI Mode, and the AI Overviews that appear on regular search results. They run as separate lanes now, at the same time.
The AI Overviews check also got sharper about reading the room. Some searches simply do not trigger an AI Overview. Now, once the results have loaded and there is clearly none on the way, the check records that and moves to the next prompt instead of waiting out the full window.
The graph learned who hangs out together
The entity graph got a real upgrade to how it finds relationships. On top of the connections it already extracts from your pages, it now derives two more layers. First, co-occurrence: entities that keep appearing on the same pages together probably have something to do with each other, and the graph now draws those edges. Second, semantic closeness: using embeddings computed on your machine, it connects entities whose meanings sit near each other even when your pages never put them in the same paragraph.
All of it runs locally against your own crawl. The graph also moved into the Tools menu, so you no longer have to remember where it lives. It needs a project with a site and a completed crawl, which is fair, since a graph of nothing is just a dot.
Moose gets live numbers
Send Moose picked up two new senses this week. He can now query the Google Search Console API for live data in the middle of a job, so a research task about your search performance uses today's numbers instead of whatever he remembered. That needs your Search Console connection, which he borrows politely and returns.
He also reads a site's llms.txt file and checks for an MCP manifest when he is sizing up a domain. He treats what he finds as context for the job, nothing more. A site that explains itself to AI is simply easier to research.
More usage on managed plans
We re-based the model pricing behind managed plans this week, anchoring every multiplier to the models' current list prices. The practical result: most models now draw 25 to 70 percent less from your monthly allowance, and several draw far less than that. Same plans, same prices, more usage, nothing to configure. One model, Kimi K3, moved slightly the other way to match its real cost, because the numbers here follow the costs, in both directions.
The small stuff
Two fixes worth a line each. Scheduled visibility runs that used only local-fetch engines were finishing their work and then never sending the summary email. The results were fine, the messenger just never left the building. That is fixed.
And if you use the audio embed to give your articles a listen button, paid plans can now remove the attribution line from the player. Free plans keep it, which seems like a reasonable trade for a free player.
FAQ
How much faster are visibility runs now?
Faster in proportion to how many engines you track, since the engines now run side by side instead of one after another. A run across five engines now takes roughly as long as the slowest lane, not the sum of all five.
Does the parallel run change what gets measured?
No. The same checks run with the same prompts and the same scoring. The only thing that changed is that they no longer wait for each other to finish.
What do the new entity graph relationships mean?
Co-occurrence edges connect entities that keep appearing on the same pages, and semantic edges connect entities whose meanings are close, based on embeddings computed locally (or optionally in the cloud using a frontier model). Together they surface relationships your pages imply but never state outright.
What does Send Moose do with Search Console access?
He queries the Google Search Console API for live data while working on a job, so research about your search performance reflects today's numbers. It uses the Search Console connection you already set up in the app.
Did managed plan prices change?
No. Prices stayed the same, and most models now draw 25 to 70 percent less from your monthly allowance per request, with several drawing far less than that. The multipliers were re-anchored to the models' current list prices.
Do free plans still show attribution on the audio embed?
Yes. The attribution line on the embedded audio player can be removed on paid plans, and stays on for free plans.
The one who's under the desk for all of it.
