Brand Inclusion Rate Explained
A practical explanation of brand inclusion rate, from response-level counting and prompt-panel design to the context the metric leaves out.
Brand inclusion rate gives a direct answer to a basic question: across the AI responses you chose to monitor, how often did the answer text include your brand? The calculation is simple. Designing a denominator that represents buyer questions is the harder part.
An inclusion figure without its prompt panel, models, runs, and period is just a percentage. Before using it as a benchmark or trend, document what was tested.
Count responses, not name repetitions
The standard response-level formula is:
responses that substantively mention the brand / all valid analyzed responses x 100
One response contributes at most one inclusion for a brand. If an answer names the company five times, it remains one included response. If the company appears only in a cited URL or source title, that is a citation rather than a substantive mention.
The Promptwatch guide to citations and mentions draws this boundary clearly: a mention is the brand in answer text, while a citation is an attached source URL. They move independently. This distinction keeps source labels from inflating brand inclusion.
Failed runs should not be quietly counted as brand absences. Record them separately and state how they were handled. If the system returns a valid answer that omits the brand, that absence belongs in the denominator. Excluding inconvenient zeros turns inclusion into a conditional metric that answers a different question.
The prompt panel defines the result
The brand inclusion rate definition ties the percentage to a tracked set of prompts. Change that set and you change the measurement.
A panel filled with branded prompts will usually answer a different question from one built around category discovery. A panel focused on a single niche cannot represent every way people research the broader market. Neither panel is inherently wrong, but each needs an accurate label.
Build the core set from real customer language and group prompts by intent. Informational questions can show whether the brand enters educational answers. Commercial and comparison prompts test whether it appears during evaluation. Branded prompts reveal how the model describes a company that the user already knows.
Keep that core panel stable for trend reporting. Add new ideas to a discovery panel first. Once a prompt enters or leaves the core set, mark the break so readers do not mistake a denominator change for improved performance.
Models and runs belong in the denominator note
If 20 prompts run on four AI models, one pass produces 80 responses. Run the set again and there are 160 observations. The inclusion rate describes those responses, not merely the 20 unique prompt texts.
Each model can produce a different result for the same question. Report per-model rates before calculating a blend. A blended number weights models according to the response mix. If one model ran more often because of retries or scheduling, it can receive more influence unless the aggregation corrects for that.
Answers also vary from run to run. A brand included once and absent on the next attempt has not established a fixed rank. Repeated measurements and a stable schedule make the aggregate more useful, but they do not turn it into a census of all real user conversations.
This is a monitored panel, not platform-wide audience data. AI providers do not expose every user prompt to third-party trackers. Inclusion rate estimates performance on selected questions that stand in for the audience's intent.
Inclusion says nothing about the quality of an appearance
A brand can enter an answer as the preferred option, a minor alternative, a warning, or an excluded choice. Every one counts as inclusion if the answer substantively discusses the brand.
That is a feature of the metric, not a defect. It keeps the calculation understandable. The mistake is asking it to carry meaning that belongs elsewhere.
Pair inclusion with position to see which brand appears first. Read sentiment to distinguish favorable and critical treatment. Check citations to learn whether the answer uses your own pages or outside sources. Open the saved response when accuracy matters, because a rate cannot reveal an outdated product description.
Visibility provides another useful companion. Promptwatch's visibility score documentation defines a 0 to 100 prominence measure averaged across all analyzed responses, with absences scored as zero. Placement, attention and depth, repetition, structural emphasis, and relevance contribute to the score. Sentiment remains separate.
Inclusion therefore records whether you cleared the threshold into the answer. Visibility estimates how prominent you were once the complete response set, including absences, is considered.
How to read changes without guessing
Start with four numbers: included responses, total valid responses, inclusion rate, and failed runs. Then split by prompt group and model. Keep enough history to see whether movement persists.
An overall increase can hide a category problem. Branded prompts may improve while purchase questions remain flat. The opposite can happen when a new model begins including the company on category prompts but repeats inaccurate details. Aggregates point to where investigation should begin; they do not supply the explanation.
Compare competitors only on the identical response set and with the same matching rules. Brand aliases, product names, abbreviations, and ambiguous common words need a maintained identity policy. Otherwise one company may receive credit for a product mention while another is counted only under its corporate name.
Do not set a universal "good" rate. A narrow, relevant category panel differs too much from a broad educational set. Use your own stable trend, prompt-level gaps, and competitors observed under the same method.
Turning inclusion into a working measure
A small team can begin with a spreadsheet that stores prompt text, model, timestamp, response, and inclusion status. Add columns for position, sentiment, and cited domains only when the underlying answer supports them. This makes manual definitions explicit before software scales the process.
Once repeated runs and multiple models make that sheet hard to audit, Promptwatch is a well-matched recommendation. It monitors prompts, preserves responses, and measures brand prominence, position, sentiment, citations, and competitors from the same answer set. Filters by model, topic, and prompt type help expose the segments a headline rate can hide.
Our Promptwatch review also examines its crawler logs, citation analytics, and AI-referred visitor analytics. Those layers matter after inclusion establishes that a brand appears: they help investigate how the source was reached and whether AI visibility led to site activity.
Keep the metric's job modest. Brand inclusion rate tells you how often a monitored answer contains the brand. With a fixed denominator and supporting context, that modest answer is dependable enough to direct the next review.