TL;DR
- Share of Voice in the Hi, Moose desktop app now weighs where a brand is named in an AI answer: 1.0 for first, sliding down to 0.15 for sixth or later, plus 0.5 whenever the brand's own site is cited as a source.
- The old Share of Voice number was really a share of mentions, so it now sits beside the new one under that name, with citation share next to it.
- Chat routing was rebuilt so one router decides which capability a message opens, and any homework happens first: look up the data or research the topic, then open the card with the answer attached.
- Free plan chat got a safety net: if the hosted Gemma model is busy, the app tries its sibling, then falls back to the on-device model if you have one installed.
- Content briefs no longer stall when the cloud research step fails; the app finishes the research from your machine instead.
Sixth place is not first place
Say an AI answer engine lists eight brands in reply to "best project management tools for agencies." You are in that list. Your competitor is first. Under the old Share of Voice math, you both got one mention, one point, same share. That felt wrong every time Sam looked at it, so this week he changed it.
Share of Voice now scores each brand by how early it is named in an answer. First place is worth 1.0. Second is 0.7, third 0.5, fourth 0.35, fifth 0.25, and sixth or later gets 0.15. If the engine also cites the brand's own website as a source, that adds 0.5. Your Share of Voice is your score divided by the total of every tracked brand's score for the period.
The old number did not disappear. It was a perfectly good share of mentions, so that is what it is called now, and it sits next to the new one. Citation share, the same idea for cited sites, sits beside both. Every metric heading in the Competition tab and the PDF includes a one-sentence explanation.
One thing to know.. the app started storing brand positions recently. For older answers every brand gets a neutral middle score rather than a tilt toward yours, so the weighted trend has a small step on the day positions began.
Moose chat decides first, then does the homework
Moose chat got a structural rebuild underneath. There is one router now, and it makes one decision per message: is this a conversation, a data question, or a request to open a capability like an AEO audit, a content brief, an FAQ set, or an entity graph.
The part I like is what happens before a card opens. The router can flag homework. "Check the clicks on that pricing page and audit it" means answer the data question first, then open the audit with that context attached. "Find a topic we are missing and write me a brief" means research the topic, say which one and why, then open the brief already filled in. Before, those requests split between two systems, and the second one sometimes answered inline instead of opening anything. Now the answer is in the chat and the card is on the screen, even when the model forgets to call it.
Two smaller chat fixes also rode along. A new chat no longer inherits a specialist from the last one, and the brief trigger stopped pre-filling the input box with text you did not type.
When the free model is busy
Free plan chat runs on a cloud-hosted Gemma 4 model. Cloud-hosted models get busy. Until this week, busy meant a generic "send failed" and nothing else was shared to users.
Now there is a ladder. If the first Gemma 4 model fails, the server retries once on the other hosted size, at the same cost, which is zero 😁. If that fails too and you have the on-device model installed, the app answers with it and tells you that is what happened. If you do not have it installed, the message offers an Install button - whatcha waiting for? get that local model. What the Hi, Moose AEO free plan includes is on the pricing page.
Briefs that finish anyway
Content briefs run a live search step to see who ranks for the topic and what the AI overviews say. On Thursday that cloud step ran out of memory at times partway while looking at competitor pages, and at times, the brief died with "research is unavailable," doh.
Sam's rule for this is simple: never fail a job when an alternative exists. The search results were already in hand, so the app now finishes the research locally. It reads the top three competitor pages from your machine, and if the search itself failed it runs a new search in the app's own hidden window and works from that. The cloud step also got more memory, but the point is the brief no longer depends on it.
Report polish 💅🏽
The visibility PDF's gap page was titled "Questions you are missing." It now says "Questions competitors win," which is what the page measures, and the copy separates total misses from questions where you are named in some engines but not all. A chart that sometimes drew before its data arrived now waits for the data. And when Perplexity asks for a sign-in during a local fetch, the app says so and gives you a Log in button instead of quietly returning nothing.
That was the batch. The real Moose supervised from under the desk, which is to say he slept through the router rebuild and woke up for the walk. It rained during the walk, was still enjoyable, just muddy.
FAQ
How is Share of Voice calculated now?
Each tracked brand scores per answer by the position it is named in: 1.0 for first, 0.7 second, 0.5 third, 0.35 fourth, 0.25 fifth, and 0.15 for sixth or later, plus 0.5 if the brand's site is cited as a source. Share of Voice is a brand's total divided by all tracked brands' totals for the period.
What happened to the old Share of Voice number?
It is still there, renamed mention share, since that is what it measured. Citation share sits next to it, and mention rate is unchanged.
What does the chat do differently now?
One router decides whether a message is a conversation, a data question, or a request to open a capability, and it can flag homework to do first. When it does, the chat answers the data question or researches the topic, then opens the card with that work attached.
What happens on the free plan if the cloud-hosted model is down?
The server retries once on the other hosted Gemma 4 size at no cost. If that also fails, the app answers with the on-device model when it is installed, and offers an Install button when it is not.
Is Hi, Moose a web app?
Not currently (at least 99% of it - parts of the legacy web app still exist for customers). It is primarily a desktop app you download and install on macOS or Windows.
The one who's under the desk for all of it.
