What every number, toggle, code and panel on the share of voice dashboard actually measures, and how to tell a real zero from nothing to measure.
One big number, five small ones, and a second figure under each. They answer two different questions and are easy to conflate.
The dashboard opens on Unbranded scope, so these are the numbers you see first.
Share of voice is Higgsfield’s citations divided by all tracked citations. Tracked means Higgsfield plus the direct and adjacent competitors named in the config, and nothing else. Reddit, YouTube and TechCrunch turn up constantly in these answers but they are logged as also-cited sources, not as brands, so they never enter the denominator. The number answers: when an answer engine names a video generation company, how often is that company Higgsfield.
The per-engine tiles are the same calculation restricted to one engine. The headline is the mean of the five, not a re-pooled total, so an engine that failed to run drops out cleanly instead of dragging the average toward zero.
Presence is the share of queries where Higgsfield is cited or named at all. It ignores who else showed up. Gemini illustrates the gap: presence 41% against SOV 9%, meaning Higgsfield gets into two answers in five but is one voice among many crowded ones when it does. Perplexity is the reverse shape, 38% presence and 17% SOV, because its answers list fewer companies each.
Presence rises when Higgsfield is mentioned anywhere. Share of voice only rises when Higgsfield is mentioned and the field around it is thin. Coverage work moves the first. Displacing competitors moves the second.
When one appears next to the headline, it compares the run you are viewing against the most recent archived run at least six days older, and the tooltip names that run and its old value. The six day floor exists so two runs fired in the same week do not produce a meaningless week over week arrow. Since the first weekly runs landed, the arrow compares each run with the one a week before it.
Every tile is also clickable. A click opens that one number’s history directly under the tile row. Section 06 covers what that chart shows.
Five controls change what you are looking at. Two of them are both called Scope.
The Scope segmented buttons under the Share of Voice heading (Unbranded / Branded / All) filter the numbers already on screen. Nothing runs.
The Scope dropdown in the status bar next to Run Now (Core / Extended / All) chooses which tier of queries the next live run will fire. It costs API calls and takes minutes. Changing it does not alter the current view at all.
Scope applies to the score tiles, Share of Voice by category, and Most-cited sources. It does not touch the query matrix, which has its own Type filter.
Classic is the flat matrix. The Data adds the per-query drill-down covered in section 05. Same underlying run either way, so switching costs nothing.
The switch in the header, beside the link to this guide. Now is everything described in sections 01 to 05: one run at a time. Over time swaps the whole page for the trend view in section 06. The Scope choice carries across in both directions.
Loads an archived run in place of the latest one. Every historical run is listed with its date, its scope, and its overall SOV, so you can pick the comparison you want rather than only seeing the most recent.
Fires a live run at the selected tier scope. Core is weekly and cheapest, Extended adds the monthly tier, All includes the quarterly probes. The scheduled weekly job runs Core on its own, so this button is for out of cycle checks.
Four panels sit between the score and the matrix. Each answers one question.
| Panel | What it shows |
|---|---|
| Coverage Map | How the 50 queries are distributed across categories and run tiers. Core 23, Extended 19, Probe 8, with a group mix line underneath. This is the shape of the measurement, not a result. |
| Most-cited sources | Every brand and publisher detected, ranked by how many answers they appeared in, blended across all five engines and filtered by the current Scope. On unbranded queries it currently reads Runway 75, Sora 48, Higgsfield 47, Reddit 47, HeyGen 42. |
| Share of Voice by category | The same SOV calculation split by query category, ranked. Unbranded scope: Reputational 12%, Commercial 12%, Film 12%, Enterprise 8%, Leadership 4%, Adjacent 0%. |
| Priority map | A grid counting queries by impact (rows, highest at top) against winnability (columns, highest at right). Top right is where a query both matters and is realistically winnable. Click a cell to filter the matrix to those queries, click again to clear. |
Every query as a row, grouped by category, with one cell per engine.
| Code | Meaning | Reads as |
|---|---|---|
| MC | Mentioned and cited | Named in the text and linked as a source. The strongest result. |
| C | Cited only | Linked as a source, name absent from the prose. The engine used Higgsfield material without crediting the brand in the answer. |
| M | Mentioned only | Named in the text with no link. Visibility without a click path. |
| – | Absent | Neither named nor cited. |
Two dots can sit alongside a code. A filled dot means the answer met that query’s win condition. A faded dot means it partially met it. Win conditions are graded on Core queries only, so an Extended or Probe row never carries a dot regardless of how good the answer was.
The second column ranks the publishers and brands cited for that query, blended across engines by default. The winner is emphasized. Higgsfield appears in the brand colour when it leads and outlined when it is in the field but not leading.
With View set to The Data, clicking a row opens the detail panel. This is where the number becomes a diagnosis. The walkthrough below uses a live row.
The panel opens with the query id and text, then the win condition and the row’s classification. For Q046:
| Query | Which companies are producing AI-generated films and series? |
| Win condition | Higgsfield named, with Hellgrind cited |
| Classification | Core · Unbranded · Film |
Five cards follow, one per engine. Here is the Perplexity card for this query, with every element labelled.
The Show full answer & reasoning button under each card opens the captured answer text plus a plain reading of how presence, share, judgment and the leader check were each decided.
Two cards can both show Higgsfield absent and mean completely different things. The sub-line is what separates them.
| Card | Share | Sub-line | What it means |
|---|---|---|---|
| ChatGPT on Q046 | 0% | not present, 2 other tracked brands appeared instead | A real zero. Runway and Luma were named, Higgsfield was not. Competitors took a field that existed. |
| ChatGPT on Q001 | — | not measurable, no tracked brand appeared in this answer | Nothing to measure. ChatGPT answered “What is Higgsfield?” with particle physics. No video company of any kind appeared, so there is no share to take. |
A dash is a name collision or an off-target answer and calls for different work than a zero, which is a competitive loss. Collapsing them into one number would hide both.
| Engine | Code | Share | Verdict | Reading |
|---|---|---|---|---|
| ChatGPT | – | 0% | fail | Runway and Luma named, Higgsfield absent. |
| AI Overviews | – | 0% | fail | Runway named, Higgsfield absent. |
| Perplexity | M | 100% | pass · positive | Named in a long studio list, no link. Leads the field. |
| Claude | M | 50% | pass · positive | Named alongside Runway, so it splits a two brand field. No leader badge, since it does not score strictly higher. |
| Gemini | MC | 100% | partial · neutral | Named and linked to higgsfield.ai/original-series. The only engine that gave Higgsfield a click path here. |
One query, five different answers, and a clear brief: Higgsfield is credible enough to be named on three engines out of five, and only one of them links.
Two summaries close the panel.
Cited leaders scoring shows how the chips in the matrix row were ranked. Each tracked brand scores 2 points if cited and 1 if only named, across all five engines, then sorts by score, then by tier, then alphabetically.
Other cited sources, mixed by type buckets every third party domain the engines linked for this query. For Q046 that is 17 distinct domains, and one line above the bar states the thing worth knowing first: Higgsfield linked directly to higgsfield.ai on 1 of 5 engines for this query.
TechCrunch and Deadline are the earned citations, Reddit and Medium the community ones, and the long neutral tail is mostly small studio sites the classifier has no entry for. Chips marked as unclassified are flagged for a human to check, not silently binned.
The dashboard re-measures every Monday. Sections 01 to 05 read one run. This section covers the view that lines runs up side by side, and how to tell a real move from normal weekly noise.
Trend lines are drawn from cadence runs only: the Monday runs, the monthly and quarterly sweeps, and the first full baseline. Manual reruns stay in the archive but are left off the lines, so a Run Now click never reads as a trend. Higgsfield currently has four cadence runs. The September 10 run was manual and is not on the chart.
| Run | Type | All | Unbranded |
|---|---|---|---|
| Sept 4 | Baseline, all 50 queries | 24% | 13% |
| Sept 7 | Weekly, Core | 23% | 11% |
| Sept 14 | Weekly, Core | 22% | 11% |
| Sept 21 | Weekly, Core | 23% | 13% |
Share of voice, blended across engines. The weekly runs refresh the 23 Core queries. Extended and Probe answers carry forward from their last sweep until the next one, so week to week, these lines move on Core answers alone.
Engine answers vary run to run with no one changing anything. The dashboard measures that from Higgsfield’s own runs and shows it beside each chart.
With four runs these yardsticks are still rough. They sharpen with every Monday.
A query’s detail panel in the Now view also carries its own history block: one strip per engine, and a line such as “present on 3 of 4 runs.” That count is the number to carry, not any single cell.
An action is anything that might move a query: an earned placement that ran, owned content that went live on a Higgsfield property, a site or technical change, or a change to how the dashboard measures. Each entry names its date and what it was aimed at: specific queries, a category, or an engine.
Every entry gets a before-and-after readout. It compares the last cadence run before the date with the first one at least three days after it, over the targeted queries measured in both, and sets the normal weekly move beside the change. It never says “caused.” Read it as what the numbers did around the date, and how large a move would need to be to mean something.
Actions also appear as numbered markers on every chart. A circle is something the team did. A square is a measurement change, and numbers on either side of a square are not directly comparable. A dashed marker tagged detected was found automatically by the dashboard run from the client's own feeds and filings (press releases, investor relations, public filings) and has not been reviewed yet; treat it as provisional until someone confirms or removes it.
The log is empty today. Click Log an action in the Over time view, fill in the prefilled template, and send it to Pete. It is added with the next update, no run needed. Name the queries it targets, since an entry without targets gets no readout. The corrected facts on /trust and /about, a new leadership page, or an enterprise listicle placement are exactly the kinds of entries worth logging on the day they land.