IQ preferences: voice, tone, and learned profile

What this covers

IQ is not a static tool. It watches how you actually respond to the decisions it surfaces, and it remembers what you tell it about your business. This article covers the three places you shape that behavior:

• Preferences, where you tell IQ which kinds of decisions to keep pushing and which to set aside, overriding what it has inferred from your own history.
• What your agent has learned, the read-out of the preferences and facts IQ has picked up about you and your operation.
• Flagging voice, the "this didn't sound like you" control on any outbound email or chat turn IQ produced, which feeds the next review of IQ's writing style.

None of this is IQ deciding anything on your behalf. It surfaces patterns from your own activity and lets you correct them. Your explicit choice always wins over what it has inferred.

Where to find it

Open Settings from the sidebar, then IQ. The IQ settings area is a six-tab dashboard at /settings/iq, with tabs Activity log, Outbound emails, Inbound classifications, Conversations, Memory, and Preferences, in that order along the top of the page under the header "IQ Activity." Visiting /settings/iq directly redirects you to Activity log; Preferences lives at /settings/iq/preferences.

The intro line at the top of every tab reads: "Every action IQ takes on your behalf, every email IQ sends, every state change in your decisions. This page is the audit you can verify against; if anything here looks wrong, that is a signal we want to see." That line applies to this whole area: it is an audit trail plus a set of controls, not a black box.

The learned-profile stats panel ("What your agent has learned") is a separate page: Settings > Privacy & Legal (/settings/privacy), under the Your Network section, below your data-contribution toggles.

Preferences: which decisions you want more or less of

The Preferences tab's header reads: "Which decisions you want more or less of," with the explanation: "Kinds of decisions you consistently pass on get set aside instead of pushed at you. If that is wrong for any of them, say so here. Your choice always wins."

How IQ decides what to set aside

IQ tracks, per kind of decision (what the product calls a signal type, for example a cash-flow flag or a certification gap), how you have responded across every one it has ever surfaced to you: acted on, resolved, or dismissed. From that it computes an action rate, the share of decisions of that kind that you acted on or resolved rather than dismissed.

If you have had enough decisions of a kind for the pattern to mean something, and your action rate on that kind is low, IQ stops pushing it and quietly parks it instead of continuing to surface it. It does not delete or hide the decisions themselves, they still exist and are visible if you go looking, it just stops actively pushing that kind at you in the feed.

The first time a kind of decision crosses into "set aside," you get a one-time notification: "I've set [that kind] aside for now," with the body: "You've passed on [that kind] a few times, so I've stopped pushing it. Want it back? Reopen it any time in Settings, under IQ preferences." It links straight back to this page. This nudge fires exactly once per kind, so it will not repeat itself every time you dismiss another one.

A section below the list of decision kinds spells out the mechanism again: "How this works. When you keep passing on a kind of decision, IQ stops pushing it and sets it aside instead. Your choices above always win: 'always surface' brings a kind back, 'always park' keeps it quiet even if you act on it sometimes."

The list

The section header above the list reads "Based on how you've handled decisions (N areas)," where N is the count of decision kinds you have any history on. Only kinds with at least one decided instance (acted on, resolved, or dismissed) appear; kinds you have never touched are not listed. Rows are sorted with the kinds you have dismissed most at the top.

Each row shows:

• The kind of decision, humanized into plain language (for example a raw internal label like cash_flow_risk reads as "Cash flow risk").
• A status pill next to the name, one of:

- Always surface (green), if you have told IQ to keep pushing this kind regardless of your history with it. - Always park (amber), if you have told IQ to stop pushing this kind regardless of your history with it. - Currently parking (amber, lighter), if you have not set an override but IQ has inferred it should set this kind aside based on your history. - Engaged (green, lighter), if you have not set an override and your history shows you consistently act on this kind. - No pill at all if there isn't yet a strong enough pattern either way.

• A summary line underneath: "N acted on · N dismissed · N% action rate," the raw counts and rate the pill above is based on.
• Two buttons, Always surface and Always park, plus a Clear button that only appears once you have set one of the two.

Clicking Always surface tells IQ: keep showing me this kind even if I have been ignoring it, useful when the truth is "I do care about this, I've just had bandwidth issues," not "stop showing me this." Clicking Always park tells IQ: stop showing me this kind even if it looks high priority, useful when the truth is "not now, I'm fully focused on something else." Clicking Clear removes your override and returns that kind to IQ's own inferred behavior based on your ongoing history.

Your override always takes precedence over the inferred pattern. If you set Always surface, IQ keeps pushing that kind of decision no matter how many times you have passed on it before or after. If you set Always park, IQ keeps it quiet even if you start acting on individual instances of it.

If you have not acted on, resolved, or dismissed any decisions yet, the list reads: "Nothing here yet. This fills in as you act on, set aside, or pass on decisions over time."

A note under the header adds: "This updates as you act on decisions. A recent change may take a day to show," so do not expect a pill or rate to update the instant you act on a decision, the underlying computation runs on its own schedule.

What your agent has learned

On Settings > Privacy & Legal, under Your Network, the "What your agent has learned" panel gives you a read-out of everything IQ has picked up about your business and how you engage with it. The section intro reads: "A snapshot of what the network has captured from you. Everything below is yours, pause contribution any time above and the corresponding rollups stop growing within 24 hours."

Three stat tiles sit side by side:

• Preferences remembered: the count of active memories IQ holds about you, things like how you like to be addressed, facts about your operation, patterns it has noticed, and past decisions. The hint line under the number breaks this down by type, for example "Preference: 4 · Fact: 3 · Pattern: 2," showing up to the top three types by count. Before IQ has learned anything, the hint reads "Will appear as the agent learns."
• Contributions to the network: the count of decision outcomes you have closed out (your feedback on what worked or did not) plus specialist claims your conversations have produced, combined into one number. The hint spells out the split, for example "12 decision outcomes · 5 specialist claims."
• Verified rollup participations: how many distinct findings in the network's anonymized intelligence layer your (anonymized) data has contributed to alongside other operators. Before you have any, the hint reads "Joins a rollup once 2 other operators independently concur"; once you do, it reads "Distinct findings the network concurred on with your data."

Below the tiles, when IQ has learned anything, a Recently learned block lists up to five recent memories, each tagged by type (Preference, Fact, Pattern, or Decision) followed by the memory content in plain language, with a confidence percentage shown beside entries where IQ has scored its own confidence, for example "confidence 82%." If IQ has not learned anything yet, this block does not appear at all.

Every number on this panel comes from your own data. There is no external join, no comparison to other operators' specifics, nothing surprising: it is a mirror of what IQ has captured from you and about you, not a scorecard against anyone else.

Flagging voice IQ got wrong

Anywhere IQ has written something on your behalf, an outbound email in Outbound emails, or its side of a conversation in Conversations, you can flag it if the tone or wording does not sound like something you would say. This is the "This didn't sound like you?" control.

1. 1Go to Settings > IQ > Outbound emails or Settings > IQ > Conversations.
2. 2Find the row for the email or chat turn that felt off, and click its flag control.
3. 3A panel opens headed "This didn't sound like you?" with the explanation: "Flag this so the next persona review can fix it. Your note (optional) helps the reviewer figure out what to encode."
4. 4Above the note field, the panel shows the context (the email subject, or the timestamp for a chat turn) and the exact content it is flagging, preserved verbatim so the flag still makes sense even if the underlying row is later edited or removed.
5. 5Optionally, type what felt off in the What felt off? (optional) field, for example the placeholder text: "used 'happy to help', sounds canned." The field has a 500-character limit; a counter under it counts down as you type ("N characters left").
6. 6Click Flag for review. The button reads "Flagging..." while it saves.

Once a flag is saved, the row is marked Flagged in place of the flag control, so you cannot double-flag the same email or chat turn, and you have visual confirmation your flag took without a page reload.

Flags do not change anything automatically. Each flag becomes a data point in the next scheduled review of IQ's writing style, where the specific phrasing or pattern you called out gets evaluated for whether it should be corrected going forward. This is a deliberately slower, batched loop, not a live edit, because voice consistency needs to be reviewed across many operators' flags together rather than patched one at a time.

If something goes wrong while saving (a network hiccup, for instance), the panel shows an error in place of the button and does not close, so you do not lose your note.

How these three surfaces relate

• Preferences shapes what IQ chooses to bring to your attention: which kinds of decisions get surfaced versus quietly set aside.
• What your agent has learned shows you the facts, preferences, and patterns behind how IQ talks to you and reasons about your business.
• Voice flags shape how IQ says things: the specific words and tone, independent of what it chooses to surface.

All three exist because Verinode is built as an independent data trust and AI Co-COO: an assistant that works for you and can be corrected by you, not one whose behavior you have to accept as given. Nothing here is Verinode deciding for you. IQ surfaces what it has inferred, and you have the final say on all of it.

Related articles

• Understanding your margin
• The decision workspace
• Acting on decisions
• The Feed
• Benchmarks overview
• How benchmarks work
• Connecting your data