Skip to content

SOCIAL SIGNALS

What the world tells you

The pages an engine reads before it answers.

Reddit threads, review sites, comparison articles and community posts are where an assistant forms its opinion of your category. This watches them continuously and ranks them by how often each one is cited in the answers you are tracking.

Someone saying something out loud into a phone, which is the conversation this module reads

What Social Signals is

Social Signals watches the off-page sources an answer engine reads before it names anyone, in the order that decides B2B SaaS answers: the review and comparison sites buyers are pointed at (such as G2, Capterra, TrustRadius and Gartner Peer Insights); Reddit and the community threads where people ask what to buy; Wikipedia and the directories assistants take entity facts from; LinkedIn; YouTube reviews; and the industry publications and comparison articles that rank for your category. For developer tools, Stack Overflow and GitHub sit near the top. It reports sentiment, clusters what people are talking about, tracks your share of the category conversation against competitors, surfaces the language your customers actually use, and writes a digest. What distinguishes it is what it refuses to keep. A mention that merely shares a word with your brand name, or that mentions you without being about you, is set aside with a stated reason you can read and overrule.

The problem

The answer was assembled somewhere you do not own.

When an assistant recommends a competitor instead of you, the reason is usually sitting on a page neither of you controls: a Reddit thread, a review site, a comparison article somebody wrote two years ago. Rewriting your own website does not touch it. The work starts with knowing which of those pages the engine actually reads for your questions, and what they say about you today.

What you get

What it reports

Sentiment and topics over time

Movement week by week with the spike explained, and what people actually talk about grouped by theme.

Explained

Share of the conversation

How much of the category conversation is yours against the competitors the rest of Signal AI Visibility Optimization (AIVO) is already tracking.

One competitor set

The words your customers use

Rarely the words in your brand book, and the earliest signal that a market has moved. It reaches the people writing, while it still matters.

Into the creative

Gaps against your own voice

Where how people describe you diverges from how your brand says it should sound, joined up so you can see the drift.

Unique to Signal AIVO

WHAT THE DASHBOARD SAYS 2,847 mentions this week +18% on last week Nobody can act on this. WHAT PEOPLE ACTUALLY SAID Onboarding is slow 412 Support response 306 Pricing unclear 221 Recommends it 198 Cost 164 The first two are a product meeting, today.
The number most tools report, and the five things behind it you can act on. Figures are examples.

Social Signals, as you would work in it

Pulse where you stand Tracking what we look for Feed what is being said Insights what it means Playbook share the findings Sentiment and volume Influencers Share of voice The words buyers use Voice gap How much people are talking, and how they feel Counts are exact. Sentiment is an estimate, so it is read as a direction. Mentions, 7 days 412 Average sentiment 0.62 Days with data 7 Mentions and sentiment, week by week wk 1 wk 2 wk 3 wk 4 wk 5 wk 6 wk 7 Mentions Sentiment
The pulse view: volume against sentiment, with the caution the product itself carries. Illustrative figures.

In practice

Volume is the easy number. It is also the wrong one.

One week of collection for a brand whose name is also a common word 54 matched a keyword Right source A theme is searched where people talk about it Not the results Financial coverage is not conversation Actually says it It has to mention what we searched for About you Sharing a word is not the same as being about you 1 genuine story 53 set aside, each with a stated reason Visible, explained, and you can overrule any of them
A real case, anonymised. Most tools would have reported 54 mentions and a healthy upward trend.

The first question anyone asks about a filter is what it took out. A filter that cannot answer that is asking for trust it has not earned, so ours shows every set-aside item with the reason attached and lets you disagree with it.

Coverage

Where it listens

Kind of source What that covers
NewsWire services and news APIs, national and regional
SocialThe major networks, plus the ones that matter in your market specifically
Forums and communitiesWhere category conversation actually happens
App storesBoth major stores, where product complaints surface first
Anything else you needAdditional sources added where a brand has one that matters

Breadth is the easy part and every tool claims it. What to ask about is the relevance test: whether the count can be traced back to the mentions behind it.

Where it connects

What the conversation feeds

  • Brand DNA gets a gap report where how people describe you diverges from how you say you sound.
  • Generated content uses the language customers actually use, taken from the mentions themselves.
  • Competitor Signals shares the competitor list, so listening and research are about the same set of companies.
  • Command Center surfaces the mentions that need a reply today.

Questions

Social Signals, answered

See what is actually being said about your brand.

Give us your brand name and three competitors, and we will run a collection before the call so you see your own mentions.

See which sources decide your answers