Key Messages

Every communications program is, at its core, an effort to make a brand mean something specific in the minds of a defined audience. Key messages are the distilled expression of that meaning — the short phrases you want people (and, increasingly, AI systems) to associate with the brand when they think about it. The Key Messages dashboard is where you capture and maintain those phrases for each property.

Why Key Messages Matter

A PR program without agreed-upon key messages drifts. Different team members describe the brand differently in different placements, spokespeople answer the same question three different ways, and the story that reaches the market is blurry instead of sharp. Explicit key messages give everyone — agency, client, spokespeople, ghostwriters — a shared reference for what the brand is supposed to stand for.

In an AI visibility context, key messages matter for an additional reason: AI platforms build their answers by synthesizing the associations they've seen repeated across the web. When a journalist, an analyst, and a podcast host all describe your client as "the most secure CRM for finance teams," that association becomes a fact the models surface when someone asks about secure CRMs. When everyone describes the client differently, no single association is strong enough to stick. Key messages give your outreach, content, and spokesperson coaching a common target.

What Goes in a Key Message

A good key message is short, specific, and ownable:

• Short: It should fit in a single sentence, ideally under ten words. If it takes a paragraph to state, it's a positioning statement, not a key message.
• Specific: "We care about our customers" is not a key message — every company says that. "The only analytics platform with SOC 2 Type II and HIPAA out of the box" is specific and defensible.
• Ownable: A key message should be something your brand can credibly claim better than its competitors. If two competitors could equally use the same phrase, it isn't a differentiator.

Key messages come in a few flavors, and it's fine to capture several per brand:

• Category definition: "The original headless CMS" — tells the audience what kind of thing you are.
• Audience fit: "Built for finance teams" — tells the audience who you're for.
• Capability claim: "Deploy in under ten minutes" — tells the audience what you uniquely enable.
• Value framing: "The fastest path from data to decision" — tells the audience what outcome you deliver.
• Contrast: "No agents, no sidecars, no headaches" — defines the brand against an alternative approach.

A typical property has three to seven key messages at any given time. More than that and the list stops functioning as a reference — nobody remembers a list of twelve.

Creating and Managing Key Messages

Navigate to Key Messages in the property sidebar. For PR-focused users, this appears under Projects because key messages most often drive campaign work. For SEO and general users, it appears under Manage. Click Add Key Message and provide:

You can edit any key message to refine the wording as the program evolves, and you can delete messages that no longer reflect the positioning. Deletions are permanent.

How Key Messages Drive Share of Influence

Key messages aren't just internal reference material — Spyglasses uses them to compute the Share of Influence (SoI) metric on the AI Visibility Funnel. SoI sits at the awareness stage of the funnel and answers a different question from Share of Voice:

• Share of Voice (consideration / decision): "When AI surfaces brand recommendations, how often is my brand named?"
• Share of Influence (awareness): "When buyers ask AI about the category — before any brand is named — how often does my framing show up in the answer?"

After every report and every nightly prompt run, Spyglasses takes each AI response and compares every sentence against every positive-sentiment key message using semantic similarity (cosine on embeddings). A sentence is a "match" when its similarity to a key message is at or above the calibrated threshold (currently 0.50) — i.e., the sentence clearly expresses the same narrative, not just the same topic.

For each report we compute, per positive key message: the percentage of awareness-stage AI responses that contained at least one matching sentence. Share of Influence is the average of those per-message rates. Three positive key messages matching in 30%, 50%, and 70% of awareness answers respectively gives an SoI of 50%.

Two practical implications:

• Negative and Neutral messages don't contribute to SoI. They're still tracked and you can see their mention rates in the per-message detail, but the funnel hero rolls up positive messages only — SoI is about the narrative you're trying to build, not the noise around it.
• Wording matters because the matcher is semantic, not exact. A sentence doesn't have to repeat your key message verbatim to count; it just has to express the same idea. That's why specific, ownable phrasings (see What Goes in a Key Message) work better than generic ones — they have a clear semantic center.

Inline Highlights in AI Reports

When you view an AI Visibility Report for a BRAND or PRODUCT property, each platform's response is rendered with inline sentence highlights drawn from key-message matches:

• Purple background — the sentence matched one of your positive key messages.
• Orange background — the sentence matched one of your negative key messages.

Hover over a highlight to see which key message it matched and the similarity score. The inline highlights are a teaser — they show up to the five highest-scoring sentences per key message per platform, so don't expect every brand-aligned sentence to be marked. The authoritative comprehensive view is the Key Messages drill-down on the Brand Dashboard, where every match is counted.

Person reports (subject type PERSON) don't use inline highlights — they retain the dedicated Tracked Attributes section instead.

Matched Answers and Sources

The Key Messages drill-down on the Brand Dashboard shows how often each message surfaces. Underneath every message row is a Matched answers link, which opens the evidence behind that rate.

The page has two tabs:

• Answers — every AI answer the message was found in, newest first. Each row carries the sentence that matched, its similarity score, the platform, the question that produced the answer, and a link to the full answer record. Filter by platform and by time range.
• Sources — the domains cited in those same answers, ranked by how many of them cited each domain. Each row shows a sample page, whether the domain is your own site, and the date that page was last updated (when we have fetched it).

The Sources tab is a correlation, not an attribution. It tells you what the AI platforms were reading in the answers where your message appeared. It does not prove that any one of those pages is where the wording came from. In practice that is still the shortlist worth working: for a rebrand, the pages still carrying the old name; for a compliance issue, the outlets repeating a claim you need corrected.

Two things to know about coverage:

• Evidence is kept going forward only. Matches recorded before this view shipped were counted but not stored individually, so an older window can show a pull-through rate with no answers listed underneath it.
• Only report and daily-tracking answers are recorded. The identity answers stored for PERSON reports are counted in the rate but have no answer record to link to.

Alerts when a claim surfaces

A key message with negative sentiment is a claim you do not want AI to repeat. Once a day, after the tracking data for a property lands, Spyglasses checks whether any of those claims turned up in the answers it measured, and tells you when one did.

An alert fires for one of two reasons:

• It is new. The claim appeared, and it had not appeared at all in the previous 30 days. Any rate is enough. A claim being repeated for the first time is worth knowing about even if it only reached one answer.
• It is sticking. The claim reached at least 20% of the measured answers on at least two of the last three days it was measured. This is the difference between a single sighting and something that keeps being said.

A claim that turns up at a low, steady rate does not trigger anything. That is deliberate: a background rate is a number to watch on the chart, not an event to interrupt someone's day with.

Each alert produces:

1. An annotation on the property timeline, dated to the day the claim was measured, naming the claim and the rate. This happens for every property, whether or not anyone is on the email list.
2. An email, to the recipients you choose, carrying the claim, the rate, up to three of the closest matching sentences with the platform each came from, and a link to the matched answers.

At most one alert is sent per claim per week, so a claim that keeps surfacing does not fill anyone's inbox.

Choosing who gets the email

Go to Settings → Reports on the property and find the Negative message alerts card. Add team members with one click, or type any address. An empty list is the default and a perfectly good setting: the check still happens and the annotation is still written, nobody is just emailed about it.

Every alert email carries an unsubscribe link. Using it removes that one address from this property's list and nothing else.

Best Practices

Make the Message Exact

The whole point of a key message is that multiple people will use the same words in different contexts. If the message is approximate — if different team members rephrase it differently each time — the repetition effect is lost. Agree on exact wording and use it verbatim in briefs, pitches, and content.

Use the Description Field for the Why

The message is what you want people to remember; the description is where you capture the reasoning behind it. Note the audience, the proof points, and the evidence that supports the claim. When a new team member comes onto the account, the description is what gets them up to speed without a two-hour onboarding call.

Revisit Messages Quarterly

Key messages aren't permanent. Competitors evolve, product positioning shifts, and what was a differentiator last year might be table stakes this year. Review the list at least once a quarter and update or retire messages that no longer hold up.

Coordinate With Projects

When you scope a campaign in Projects, reference the relevant key messages in the project brief. This keeps the campaign narrative tied to the brand's overall positioning rather than drifting into one-off angles.

Don't Confuse Key Messages With Taglines

A tagline is customer-facing — it shows up on the homepage. A key message is internal-facing — it shows up in briefs, pitch notes, and spokesperson training. The same phrase can sometimes be both, but most key messages are written for working teams, not for billboards.

Track message drift with an AI assistant (MCP)

Connect the Spyglasses MCP server and an assistant can narrate how your messaging lands over time. get_message_tracking returns each key message's pull-through rate into AI answers, and get_answer_summaries returns the full weekly answers for a tracked query so the assistant can describe exactly how the AI's wording drifts week to week. The Track message drift prompt chains both.

See MCP → Account data for the full tool reference.

Related

• MCP Server — Track message pull-through and drift from an AI assistant
• Projects — Scope campaigns around the key messages you want to reinforce
• Annotations — Mark campaign milestones on your visibility charts
• Tracking Awareness — See how key messages drive Share of Influence at the awareness stage
• Prompts — Author the awareness-stage prompts that feed Share of Influence