Reputation AI agents

What this is

Reputation isn't read by one generic AI. Behind the section sits a small bench of specialists, each one built for a specific job in the pipeline that turns raw reviews and scorecards into something you can act on. One clusters your open reputation signals into decisions in real time. One turns a decision into a full 30/60/90-day recovery plan. One runs a nightly sweep looking for patterns no single review would show on its own. And a fourth, scoped to franchise HQ leadership rather than an individual operator, reads the same kind of signal across an entire network.

This article covers what each of the four does, in what order they hand off to each other, and the agent activation tile, the single click at the top of Reputation's Take Action row that opens a conversation with IQ and starts the section's autonomous work going forward.

For the rest of the section, start with the Reputation section overview. For how a decision these agents produce actually reads and behaves on screen, see Reputation Findings and decisions.

Where to find it

Open Reputation from the sidebar, at iq.verinode.ai/reputation. You never talk to these four agents by name or pick one from a list. They work behind three places on the page:

• The Take Action row, where the recovery strategist's plans and the decision orchestrator's clustered verdicts show up as decision tiles.
• The Findings tab of the Reputation card slider, the full gallery of every open decision these agents have produced.
• The agent activation tile, first in the Take Action row, your one click into a conversation with IQ scoped to Reputation.

The network reputation analyst doesn't live on /reputation at all. It's an HQ-side specialist, reachable from the network's own Co-COO chat, covered near the end of this article.

The pipeline, in order

Three of these four agents form a pipeline. Data flows in, gets clustered into decisions, and one of those decisions can become a full recovery plan. Here's the order they actually run in:

1. 1Detectors fire first. Verinode's per-platform detectors read your reviews, response times, and scorecards and write individual signals: a score drop on Google, a cluster of negative reviews with a shared theme, a carrier↔customer divergence, an unclaimed profile. These are the atomic facts, not yet a decision.
2. 2The Reputation Decision Orchestrator clusters them. Every time detectors run, the orchestrator looks at your open signals and groups the related ones (a negative cluster plus a slow response time on the same platform, a score drop that lines up with a carrier scorecard gap) into up to three coherent decisions. Single, unrelated signals stay as plain alerts; only clusters become decisions.
3. 3A verdict gets picked and handed off. Each decision the orchestrator writes carries exactly one verdict, and the verdict decides which specialist executes it next (see "The five verdicts" below).
4. 4The Reputation Recovery Strategist executes the strategic verdicts. When a decision's verdict is recover or claim, it's this specialist that writes the actual 30/60/90-day plan you see once you click Act.
5. 5The Reputation Synthesizer runs nightly, independent of the above. Once a night, it re-reads your whole reputation posture (per-platform standing, customer surveys, carrier scorecards, the carrier↔customer gap, checklist progress, and whatever signals are still open) looking for compounding patterns the orchestrator's real-time pass might not connect. When it finds one, it writes its own decision, tagged so it can be told apart from an orchestrator verdict.

None of this replaces your judgment. Every decision these agents produce is a recommendation with a dollar figure or a risk statement behind it. You read it and choose Act, Not now, or Ignore, exactly as Reputation Findings and decisions describes.

The Reputation Decision Orchestrator

What it does. Runs every time the per-platform detectors fire (part of Verinode's nightly pipeline). It reads your open reputation signals and groups the ones that belong together into up to three decisions, then picks a verdict for each one.

The five verdicts. Every decision the orchestrator writes carries exactly one of these:

• Recover. A multi-signal cluster needs a full 30/60/90-day plan. This is the default whenever a cluster spans more than one platform, or compounds a response-time problem with a quality one. Routes to the Reputation Recovery Strategist.
• Respond. A tactical cluster of unresponded reviews on a single platform. This doesn't need a strategy, it needs replies drafted, so it routes to the review-responder specialist instead.
• Claim. You've added a profile Verinode hasn't been able to scrape yet, usually because it was never actually claimed. Routes to the Recovery Strategist for the connect-and-monitor playbook.
• Defer. Not enough evidence yet to act. The orchestrator uses this sparingly, typically only when the signals behind a would-be cluster are under two weeks old and nothing else in your posture looks unhealthy.
• Accept risk. An explicit "we know about this and we're choosing not to fix it right now" verdict, used only when you've already signaled that intent, for example by dismissing a related decision with a reason.

What backs each verdict. The orchestrator never picks a verdict blind. Internally, it weighs at least two options for every decision, for example considering "recover" against "respond" and explaining why a full strategic plan beats just drafting replies for that particular cluster, or the reverse. You don't see that weighing directly, but it's why a decision's framing reads as a reasoned call rather than a template.

Dollar figures. When a decision carries a cost-of-inaction number, it's grounded in your own data (your lead volume, your job volume through a specific carrier program), never a generic industry guess. If your own numbers don't support a figure, the orchestrator leaves it off rather than estimating one, so an untagged decision means exactly that: no dollar figure could be justified yet, not that the risk is small.

How urgency is set. Every decision also carries an urgency window, in days. An unresponded negative review sits at the short end of that range. A score drop confined to one platform sits further out. A carrier↔customer divergence, where the carrier's own scorecard hasn't caught up to what customers are already saying, sits furthest out, because the clock only really starts ticking once the carrier notices.

The Reputation Recovery Strategist

What it does. Takes a recover or claim decision and turns it into the actual plan you work from once you click Act on a decision tile. This is the specialist behind the checklist you see inside a decision's action-plan workspace at /decisions/[id].

The plan has six phases, and not every phase appears on every plan:

• Phase 0, Diagnose (this week). A short read on what's actually driving the decline, naming which dimensions are compounding (the negative-review theme, the response time, the platform, the direction of any divergence). If your own review volume is too thin to call it a real trend yet, the strategist says so directly and recommends holding off on a full plan in favor of something smaller, like wiring up review notifications.
• Phase 1, Stop the bleeding (30 days). The near-term fixes: getting review notifications wired so a negative review doesn't sit overnight, setting a first-touch reply target, and naming the specific outstanding unresponded reviews (by count and platform) that need drafts, which it routes to the review-responder specialist rather than drafting itself.
• Phase 2, Address the root cause (60 days). The systemic fix behind the symptom, whether that's a communication cadence, a documentation mismatch between what a carrier sees and what a homeowner experiences, a capacity or staffing issue, or a pricing theme that gets routed to the competitive-analyst specialist for carrier-program-rate context.
• Phase 3, Rebuild reputation (90 days). How you generate positive signal to offset the recent decline: a post-job review-request cadence you run through your own existing channel (Verinode does not message your customers on your behalf), wiring in a CRM's post-job survey data if you already collect it so the survey-analyst specialist can read the themes, closing gaps in your profile completeness, and, when your carrier scorecards are healthy, leaning on that standing in your public response copy.
• Phase 4, Carrier-side coordination. Only appears when a carrier↔customer divergence signal is part of the cluster. The move here is proactive: brief the carrier program contact on the customer-side trend before their own scorecard cycle catches it, framed as "here's a gap we're already closing," not a defensive response to a complaint.
• Phase 5, Track progress. What to watch going forward, your trust score trend on the affected platform, whether the negative-review cluster is shrinking, whether your median response time is holding under 24 hours, and whether your customer survey response rate is climbing.

What it never does. It doesn't draft the individual review replies itself (that's review-responder's job) and it doesn't interpret survey results on its own (that's survey-analyst's job). It sets the strategy and names exactly which specialist executes each piece.

The Reputation Synthesizer

What it does. Runs once a night as part of Verinode's nightly pipeline, independent of the real-time orchestrator pass. Where the orchestrator reacts to whatever signals just fired, the synthesizer steps back and reads your entire reputation posture at once, every connected platform's stance, your customer survey aggregate, your carrier scorecards, any carrier↔customer divergence, your checklist progress, and every signal still open, looking specifically for patterns that only show up when two or more of those dimensions are read together.

What counts as a stack. The synthesizer never surfaces a single dimension on its own, that's already covered by a per-platform detector or the orchestrator. It only writes something up when at least two dimensions compound into one story. A few examples of what that looks like in practice: a negative-review cluster about response time, showing up alongside your own response-time detector and a low customer survey response rate, is a three-layer customer-experience story. A carrier↔customer divergence combined with a Google rating drop in the same window is a "the wedge is opening before the carrier's own numbers show it" story. A decline across more than one platform combined with a mostly-incomplete reputation checklist is a "the fundamentals haven't been set up yet" story.

What each stack contains. A one-sentence headline naming the pattern, two to three sentences on the mechanism behind why these particular things compound (not just a list of what's true), a root cause when a single shared driver exists (or none, when the pattern is structural rather than caused by one thing), and an ordered 30/60/90-day action sequence where every step names which specialist should be consulted next: the Recovery Strategist by default, review-responder for tactical reply drafting, survey-analyst for reading customer survey themes, or competitive-analyst when a carrier-program gap is part of the picture. A dollar figure appears when the synthesizer can ground one in your own numbers, and is left off otherwise.

Where it shows up. A synthesizer stack surfaces in the same Take Action row and Findings tab as an orchestrator decision, since both flow through the same decision pipeline described in Reputation Findings and decisions. It's capped at three stacks per sweep, and it stays silent entirely on a night where nothing actually compounds, which is the normal, healthy case for most operators most nights.

The agent activation tile

What it is. The first tile in Reputation's Take Action row, before any decisions load in. It's a single click that opens IQ's chat panel on the right side of the screen with a Reputation-scoped conversation already seeded, so starting a conversation about your standing costs exactly one click, not a blank text box you have to know how to fill.

What you see. The tile carries the label Start here, the headline Talk to IQ, and this thesis line underneath: "I watch your online presence, review trends, response gaps, sentiment shifts that show up before customers stop calling." An Ask IQ chip sits in the corner with an arrow.

What happens when you click it. The agent panel opens (un-minimizing if it was collapsed), and if you don't already have a conversation going on this route, IQ's first message seeds in automatically:

"I'm IQ. I watch your online presence, review trends, response gaps, sentiment shifts that show up before customers stop calling. Right now I have nothing to read. The fastest way in: forward a review notification or paste in your Google Reviews link. Once anything lands I'll start tracking sentiment trends, flag reviews that need a response, and watch how your overall rating trends over time. Your peer comparison fills in as more operators in your area join. What's your biggest reputation worry right now, a recent bad review, or just radio silence?"

That last question is deliberate. IQ qualifies what you actually need before it recommends anything, rather than guessing.

Why it's called an activation tile. This is the moment the section's autonomous work actually has something to build on. The detectors, the orchestrator, and the synthesizer all run on a nightly cadence regardless, but a conversation is what turns a cold section (no platform connected, nothing to read) into one where IQ has real context to reason from going forward: a starting question you asked, a document you forwarded, a worry you named. From here on, the Recovery Strategist has your actual situation to work with instead of a blank profile.

Once you've used it, it's gone. The tile retires permanently after your first engagement, whether that's clicking it directly or simply having any conversation history on the /reputation route from a prior session or another device. It doesn't come back once dismissed, and it doesn't count against or replace any of the actual decisions that show up in the row after it, those are two separate things: the tile is your entry point, the decisions are Verinode's output.

The Network Reputation Analyst (HQ)

What it does. This one isn't scoped to a single operator. It's part of the Verinode | HQ Co-COO's bench, serving franchise HQ or association leadership who need to read reputation across an entire network of locations rather than one business.

How it frames things. Always as the network, "N of M locations," never calling out one franchisee by name. Its job is to tell HQ whether a reputation dip is a brand-wide issue (a policy change, a systemic service gap touching most locations) or a handful of local problem spots while the rest of the network holds steady, because those two situations call for different plays from HQ. It treats reputation as a leading indicator: a slide today shows up as softer lead volume weeks later, which is why it frames the stakes in terms of acting before the number becomes a crisis rather than after.

What it recommends. A recovery play HQ can push down, usually some combination of a review-response cadence across affected locations, a service-recovery step for the specific low scores driving the pattern, and sharing what the network's own best-reviewed locations are doing differently, grounded in this network's actual top performers, not a generic best practice.

Where to find it. It surfaces through the HQ Co-COO's chat when a network-wide reputation trend signal routes to it, and its output points back to the HQ Reputation surface for the underlying detail. If you're an individual operator reading this from /reputation, this agent isn't part of your experience; it belongs to the separate HQ product built for network leadership.

Best-practice example

Say your Google rating has slipped over the past two weeks and three recent reviews all mention slow scheduling. The per-platform detectors pick up the score drop and the negative cluster as separate signals. On the next pipeline run, the Decision Orchestrator sees both signals point at the same platform and the same underlying theme, clusters them into one decision, and picks recover over respond because the pattern spans more than a single unanswered review. You open the decision tile and click Act. The Recovery Strategist takes over: Phase 0 confirms the volume is real, not noise; Phase 1 tells you to wire up review notifications and lists the specific outstanding reviews that need replies; Phase 2 points at your scheduling cadence as the root cause to fix; Phase 3 lays out a post-job review-request habit to rebuild volume; Phase 5 tells you what to watch. Two weeks later, the nightly Synthesizer notices your survey response rate has also been thin the whole time and your reputation checklist is still under half done, and writes up a second, broader stack connecting all three, with its own action sequence naming the survey-analyst specialist for the piece the Recovery Strategist's plan didn't cover.

Related reading

• The Reputation section overview
• Reputation Findings and decisions
• The decision workspace
• Forwarding documents
• Connecting your data

Data sources