SignalOS features
SignalOS features are the 53 capabilities in the product today, in five groups. They measure how ChatGPT, Claude, Perplexity and Gemini describe your brand, diagnose why you're missing, act on the gaps, prove what changed, and run it as a service. Estimates are tagged as modelled, and what isn't built is listed at the end.
What do the markers beside each feature mean?
Which plan includes each feature is shown on pricing. Plan markers appear beside a feature once plans are published.
"Core" marks the features the rest of the product depends on. "Modelled" marks an estimate built from a formula and your own assumptions rather than a measurement, and the product labels those as modelled wherever they appear, including in exports. Our methodology explains the difference in full.
A feature marked "Every plan" isn't limited to particular plans. Some features are, and for those the marker names the plans that include them. It's read from the live plan catalog each time the page loads, so it can't disagree with what you'd actually be sold.
If you'd rather see the features working together than read them as a list, how SignalOS works follows one brand through a full cycle. The API and MCP server expose the same data, with the same confidence and sample size shown here.
What does SignalOS measure?
Everything downstream depends on the measurement being defensible, so this is where most of the engineering went. SignalOS asks the questions your buyers ask, repeatedly and in clean sessions, and records exactly which model answered.
Multi-engine measurementEvery plan
ChatGPT, Claude, Perplexity and Gemini are each measured separately. No engine is held back for a higher plan.
Repeated samplingCore · Every plan
We ask each question five times per engine per cycle. Presence is reported as a rate with a confidence interval, not as a yes or no from a single answer.
Sampling that degrades honestlyEvery plan
If your budget or measurement volume runs low, a run takes fewer samples per question rather than dropping an engine, and never fewer than three. Below that, the engine isn't sampled for that run, and you're told which engine, why, and when it resumes.
Model-version trackingEvery plan
Every answer records the model version the provider reported. When a provider changes models, the change appears as a marker on your trend charts.
Clean-session runsEvery plan
No memory, no prior turns, no personalization. An engine that can't guarantee a clean session is refused rather than sampled.
Entity resolutionEvery plan
An alias registry covers legal names, abbreviations, misspellings and product names. New aliases are only suggested until a person approves them, and ambiguous matches go to a human review queue.
Mention classificationEvery plan
Not just whether you appear: your rank among the options named, the sentiment, the segment the answer places you in, and the attributes it attaches to you.
Query universe managementEvery plan
Questions are clustered by topic and intent and tiered by how they actually perform. A refresh proposes additions and retirements for a person to review before anything changes.
Personas and geographiesEvery plan
Measure the same questions as a different buyer or from a different country. The estimated cost of the extra measurement is shown before you confirm it.
Volatility scoringEvery plan
How much a question's result swings from run to run, banded low, moderate or high. A high rate that's stable is a different situation from one that moves.
How does SignalOS find why you're missing?
Absence has causes, and they need different fixes. SignalOS separates a crawler block from a thin citation footprint from a page no engine can extract an answer from, rather than reporting one number.
Citation source analysisCore · Every plan
The sources each answer was built from, how much each shaped it, and who owns them: you, a competitor's own site, or a third party.
AI crawler access auditEvery plan
Checks your robots.txt against twelve named AI crawlers, including GPTBot, ClaudeBot, PerplexityBot and Google-Extended, and says what each one blocked means.
Retrievability scoringEvery plan
Grades each page on seven signals: a direct answer early, question-shaped headings, self-contained sections, entity definitions, dated figures, comparison tables and schema. Every finding quotes the real text it's about.
JavaScript-dependency auditEvery plan
Compares a page's raw HTML with the rendered page to measure how much of its content a crawler that doesn't run JavaScript never sees.
Technical auditEvery plan
Status codes, noindex, titles, meta descriptions, canonicals, H1s, HTTPS and duplicate titles, plus Core Web Vitals captured in a real browser for the pages you choose.
Entity and schema auditEvery plan
Whether your structured data is valid and your entity details agree across your own pages, and whether Wikidata recognizes you at all.
Content auditEvery plan
Tracked questions no page appears to cover, pages competing for the same question, and pages we haven't re-crawled in six months. Matching is literal, and the audit says so.
Competitor trackingEvery plan
Competitors are classified by exactly the same rules as your brand: the same alias matching, rank, sentiment, segment and attributes. No asymmetry.
Share of voiceEvery plan
Weighted by how prominently each brand appears and by question quality, not a raw count of mentions. Where the sample is large enough, each brand's score carries its own confidence interval.
Findings with a lifecycleEvery plan
Eleven states from detected to archived, deduplicated by fingerprint so a repeat issue isn't a new one, with regression detection when something resolved comes back.
What does SignalOS help you do about it?
Diagnosis without the work is a report, not a service. SignalOS turns each gap into specific work, ranked, sized to your capacity, and checked against what a client has said they won't do.
Citation gap worklistCore · Every plan
Every cited source, bucketed by who owns it and whether it shaped the answer. A play is suggested only where one genuinely exists, and a competitor's own site never gets one.
Outreach plays
For third-party publishers the engines rely on where you're missing. Each play is prepared, then tracked through contact to placement.
Review platform plays
Targets your own profile on the review platforms the engines cite, such as G2, Capterra and Trustpilot.
Entity remediation
Wikidata work, with a notability assessment first, because an entry that gets rejected is worse than none.
Original research plays
When no good source exists for a question, becoming one is the strongest move. The play comes with a brief for the research piece.
Content brief generator
From any opportunity: the target questions, the entities to cover, a structure built on the seven retrievability signals, internal links, and sources worth citing.
Opportunity scoringEvery plan
Business value, search opportunity, competitive gap, expected impact, confidence and effort, each visible. Weights are set for the organization, can be overridden per client, and every opportunity keeps the weights it was scored with.
Strategic constraintsEvery plan
Record what a client won't do. A constraint scoped to a market or content type blocks matching recommendations, which are listed with the reason rather than silently dropped; a free-text constraint raises a warning.
Capacity-paced worklist
Declare the hours you can absorb and the worklist is sized to them. The rest waits in an ordered backlog, and work blocked by a prerequisite is held without using capacity.
Compliance review
Flags regulated-sounding claims such as HIPAA, GDPR or certification language for a person to review, with every decision audit-logged. It's a review queue, never a compliance verdict.
How do you prove what changed?
A number that moved is only useful if you know why. Each of these is built to survive a client asking "how do you know", and the statistics behind them are written up in our methodology.
Evidence drawerCore · Every plan
Click a number and read the raw AI answer behind it, with your brand mention highlighted, the model version, the run timestamp and the distribution across runs.
Impact analysis
What changed after a piece of work, read through a lag model per type of work, because a citation earned today doesn't change an answer tomorrow. A reading taken too soon has its confidence downgraded rather than counted as proof.
Confounder flagging
When a model version changed, another piece of work overlapped, or an external event was recorded in the same window, the impact reading says so rather than claiming the credit.
Score attributionEvery plan
A change in visibility broken into named causes (content published, new citations, a competitor gaining ground), with the unexplained residual always shown. The bars add up exactly.
Visibility-to-value modelModelled · Every plan
An estimate of what a cluster's visibility is worth, built from your own deal value and conversion assumptions. It's labelled modelled everywhere, and withheld when the sample is too small.
Forward simulationModelled
What a planned change might produce, projected from your own past results. Below five completed impact analyses it refuses to show a number at all, because there would be nothing to base one on.
Published methodologyEvery plan
Sampling, thresholds and model versions, written up inside the product. Classifier accuracy is measured against human-labelled examples, reported as unmeasured below twenty, and shown in full to roles with evaluation access.
Methodology versioningEvery plan
A change to how numbers are computed can't ship without a version bump and a dated changelog entry, and each change is marked on your trend charts.
What do you need to run it as a service?
The parts that matter once AI visibility is a service line rather than a curiosity: your brand on it, your clients inside it, and your own systems reading from it through the API.
White-label
Your logo, favicon, colors, typography, sender name and footer, with per-client overrides. It hides our name in the client portal, presentation mode and PDF exports, and a color that would fail contrast is rejected.
Client portal
Five screens in plain language: overview, trend, this period, what's changed and what's next. Each has a comment thread your client and your team share.
Pooled measurementEvery plan
One measurement pool for the organization, with what each client, engine and cluster consumed ranked so you can see where it went.
Per-client caps
Set a planning cap for each client inside the pool. Overallocation is allowed deliberately and shown clearly, because not every client uses its full cap.
Roles and permissionsEvery plan
Six roles: owner, admin, strategist, operator, viewer and client user, enforced on the server. Strategists and operators get dashboards built for their work, and client users land in the portal.
AlertingEvery plan
Events and movements are treated differently: a movement has to pass a statistical significance test before it alerts, so random fluctuation doesn't wake anyone. Alerts can go to Slack, Teams or a webhook URL.
Ask SignalOSEvery plan
Plain-language questions answered by fixed queries over your real data. A model may pick which question you meant; it never produces the figure, and too little data gets a refusal.
ReportsEvery plan
Branded PDF reports with every modelled figure marked, so an estimate can't pass for a measurement once it leaves the product.
Scheduled reports
The same reports on a recurring schedule. A schedule whose creator no longer has access fails visibly rather than running under someone else's name.
API and MCP server
Every API response carries a provenance object with the confidence, sample size and a link to the evidence. The MCP server answers from the same fixed queries as Ask SignalOS, and a key stops working after a downgrade.
WebhooksEvery plan
Signed event deliveries to your own systems. Today they cover new tracked questions and new chart annotations, and subscriptions can be paused and resumed.
Presentation and wall modeEvery plan
Any dashboard becomes full-screen slides for a client meeting, and a rotating board of every client's visibility runs on an office screen. Same data, no chrome.
Usage transparencyEvery plan
What you've consumed, a projection to the end of the period, which client, engine and cluster consumed it, and the estimated cost of a run before you start it.
Provider selection
Turn off a specific AI engine for one client when their contract or data-processing terms rule it out. A restriction already in place keeps applying after a downgrade.
SSO and SCIM
Single sign-on over SAML 2.0 or OIDC, inactive until an owner turns it on, and SCIM provisioning to add and remove users from your identity provider.
Not built, and we'll say which
A feature list that only says what a product does leaves you to discover the rest after you've bought it. These are the things people ask about that SignalOS doesn't do today.
Google AI Overviews
There is no compliant public API for AI Overviews or AI Mode, and we won't scrape search results to get around that. Covering it would need a licensed data provider.
Automatic placement monitoring
When a play earns a placement, checking whether it now shows up in AI answers is a manual step: a person records what they found. Nothing watches for it automatically.
Non-English measurement
Our classifiers read English. A non-English answer is marked as excluded and left out of your rates, never counted as your brand being absent.
Revenue attribution
We estimate what visibility might be worth and label that estimate as modelled. We don't measure revenue from AI visibility, and nobody has shown that it can be measured.
Server log analysis
There's no ingestion or analysis of your server logs. It was once listed on the Enterprise tier without being built, and was removed from it on 2026-09-14.
Customer-facing audit log export
There's no dedicated audit log export for customers. Your organization's audit log is included in the full data export every organization can request. The dedicated export was once listed on the Enterprise tier without being built, and was removed from it on 2026-09-14.
Common questions about SignalOS features
- Which features are on which plan?
- Every plan measures and diagnoses. Acting on the gaps, reselling the service under your own brand, and governance controls such as SSO are added as plans go up. The markers on this page are read from the live plan catalog, and pricing shows the full comparison.
- What does "modelled" mean?
- A modelled figure is an estimate calculated from a formula and your own assumptions, such as deal value and conversion rate, not something we observed in an AI answer. The product labels it as modelled everywhere it appears, including exported reports, and our methodology lists which figures are modelled.
- Can I turn features off?
- There's no single switch for each feature, and most of them only act when a person starts something: a play, a brief, a report or a schedule. White-label is opt-in, and an individual AI engine can be turned off for one client where a contract rules it out.
- How often do you ship?
- Continuously, without a fixed release calendar. Changes to how a number is computed are the exception that gets ceremony: each one requires a new methodology version and a dated changelog entry, and is marked on your trend charts, so a change in our method is never mistaken for a change in your visibility.
See which of these matter for your brand.
A walkthrough runs against a brand you actually care about, and shows which of these features are relevant to it and which aren't.