Voice review

Talk through the deck

The agent holds the full text of all nine deck sections, the project summary and every decision made, the file index, and last night’s complete voice transcript. Ask it anything about the deck, or just talk and it records a brain dump. It is read-only and cannot approve or execute work. Allow the microphone when your browser asks.

The explicit Sonnet rollback keeps the existing capsule-only protocol. A lost connection retries the same Sonnet engine twice and reloads the complete persisted conversation. It never changes models during a conversation.

Sonnet 5 · no fallback

Microphone Using: browser default until voice starts

offline

Live transcript

not started

Voice answers are review evidence. Approval, sending, spending, and production changes stay in the originating chat. Voice trouble in this page? Open the standalone voice page.

Voice companion for the Fusion Tables PPC deck

Private. Talk through the deck out loud instead of reading it twice.

What this agent already knows

The complete readable text of all nine deck sections, S1 to S9, exactly as published, in Hungarian.
A private project summary covering the client situation and every decision already made, including the answers to Q1 through Q6 that shaped the deck.
The full transcript of the voice session from the night of 2026-08-10, 176 turns, the one that produced this deck.
A file index naming every draft, data pull, QA round and screenshot on disk, plus the canonical figures and the figures that must never be used.
The action plan with its clarifications, PROJECT_STATE.md, and the changelog.

It cannot open files mid-call. When a question needs something only on disk, it names the exact path so you can open it yourself. It cannot approve, send, deploy, spend or change anything.

Things worth asking it

WalkthroughTake me through S1 section by section, shortest version.

Trace a numberWhere does 129 418 Ft come from, and what window is it?

Weak spotsWhat is the least defensible claim in the whole deck?

CompetitorsWhy Gamecenter and 4Biliárd, and who got rejected?

The illustrationsWhich panels are mock, and is the labelling honest enough?

Client readRead S2 back to me as Márta would hear it.

Last nightWhat did I say in the voice session that did not make it into the deck?

Brain dumpJust talk. It records change requests read-only.

The nine sections

S1 Current ads, analysed ad by ad, with per-ad metric tables for Meta and Google.
S2 Measurement assessment. The section that found the broken step.
S3 Dashboard. Real measured numbers, plus clearly labelled illustrations for what is not measurable yet.
S4 The next campaigns: targeting, ads, landing pages, A/B tests.
S5 Competitors, two Hungarian and two American.
S6 Keyword table.
S7 Category launch order and media split.
S8 CRM upgrade plan for the Google Sheet.
S9 Google Business Profile and reviews, light pass.

Open the deck Build scoreboard Action plan review

Afterwards

Nothing said here changes the deck by itself. To act on the conversation, the session transcript has to be imported back into the coding session, which then edits the fragments, re-runs QA and republishes the same slug.