Skip to main content

​
What is a topic?

A topic is a named group of prompts that belong to the same buying question. You assign a prompt to a topic, and Attensira rolls the per-prompt numbers up into a single set of rates for the group. Nothing about how a prompt runs changes — topics are a reporting structure, not a scheduling one. The reason to use them is statistical as much as organizational. One prompt on Starter gives you one reading a day; ten prompts in a topic give you ten. A single prompt swinging from 0% to 100% is often noise. The same swing across a topic is a signal you can act on. In the wire format a topic is group_id, but the get_analytics area enum value is topic, and POST /v1/prompts files a prompt under one with a topic field. The published vocabulary and the enum agree; only the identifier field kept its original name.

​
What changes when I group prompts?

Reporting rolls up. Instead of reading forty prompt rows, you read six topic rows, each carrying a share of voice, a citation rate, and competitor rates computed across every successful run of every prompt in the group. That rollup is a pooled rate, not an average of averages. A topic with two prompts — one that produced three successful runs and one that produced nine — weights the twelve runs together. This is what you want: the topic answers “across the questions in this theme, how often were we named?” Individual prompt rows do not disappear. Grouping adds a level of reporting above them; you can still open any prompt and see its own rates, its citations, and the source domains the models drew on.
A prompt without a topic still runs and still reports. Topics are optional. Prompts left ungrouped appear only at the prompt level.

​
The starter set

A new workspace does not open on an empty grid. At signup the agents choose its first prompts and file them into named topics, so by the time you reach the competitors step it already tracks about fifty prompts in four to eight topics. On the goals step they are shown collapsed: one line with the count, opening to the topics and then to the prompts. The prompts come from three sources, and the order is the precedence:
  1. The free audit’s prompts, word for word, if you came from the free audit. Rephrasing them would make the product’s first number disagree with the one that brought you here.
  2. Prompts suggested by the analysis of your site.
  3. New prompts the agents write from your business profile. These are phrased the way a buyer asks an assistant, 4 to 14 words, and never name a brand, yours or anyone else’s: an unbranded question is the honest test of whether a model reaches for you unprompted. Topic names never name a brand either.
Duplicates are removed, and the set is capped at the prompt slots you have left: your plan’s, or Starter’s 50 before you hold a plan, less the prompts your other workspaces already track. When the cap bites, the audit’s prompts are kept first and the agents’ own are trimmed evenly across topics. A chosen prompt that fits none of the proposed topics goes into a topic called More questions. When the agents cannot tell what your business does from your site and profile, they do not guess. The starter set is then the audit’s and the site analysis’s prompts only, in one topic called Starter questions, and the goals step says why. Fill in your business profile and they can propose more. The starter set is created once. After that it is an ordinary set of prompts and topics: add, remove, reword or regroup any of them, from the prompts screen, chat or MCP.

​
How should I structure topics?

Group by the question a buyer is asking, not by the feature you want to sell. Buyers arrive at assistants with a job to do, and the topic that predicts your outcome is the job, not your roadmap. Three structures that hold up:
  • By stage. Discovery prompts (“what tools exist for X”), evaluation prompts (“X vs Y for a team of 20”), and objection prompts (“is X hard to migrate to”). These behave very differently — discovery answers name many brands, evaluation answers name few.
  • By segment. The same question asked for a solo operator, a 20-person team, and an enterprise. Models often recommend entirely different products across these, and a blended number hides it.
  • By use case. One topic per problem your product solves. This is the structure that maps most cleanly onto content work, because a weak topic points at a page you need to write.
Keep topics large enough to be stable — five to fifteen prompts is a workable band — and resist a topic that holds one prompt, which reports exactly what the prompt already reported.

​
How do I create and change topics?

1

Create the topic

Use New topic on the Prompts or Topics screen, or let the agent propose topics when you paste prompts. Name it in the words a buyer would recognize. The name appears on every rollup row, so make it readable without your internal shorthand.
2

Assign prompts

A prompt belongs to one topic at a time. Assigning it to a new topic moves it rather than copying it. File prompts as you add them, move one from its own page, move many with the bulk bar, or drag them between groups (below).
3

Read the rollup

Topic rates appear once the prompts in the group have successful runs. A brand-new topic reports value: null until then.
Regrouping is retroactive, and it takes effect at once. Nothing is stored per topic: every topic number is computed when you read it, from the stored runs and the topic each prompt is filed under now. Move a prompt and the next read of either topic counts that prompt’s whole history in its new topic, with nothing to rebuild and no wait for the next reading. The same holds everywhere a topic is a cut: the boards, sentiment by topic, query_evidence grouped or filtered by topic, and answer_slice filtered by topic. Historical topic charts will therefore shift when you reorganize. Do a reorganization in one sitting rather than drifting prompts between topics week to week, or your trend lines describe your bookkeeping instead of your market. Deleting a topic does not delete its prompts. They become ungrouped and keep reporting individually.

​
Merging, splitting and dragging

The structure the agent proposes, or the one you drift into, rarely stays right. Three reorganisations fix it without re-adding anything:
  • Merge into… is for two topics that turned out to be one question. Every prompt of this topic moves into the topic you choose, with every reading, this topic is deleted, and you land on the topic you merged into.
  • Split… is for one topic that turned out to be two. Name the new topic and tick the prompts that go to it; the rest stay. The new topic takes the engines and the country of the one it came from. A split needs at least two prompts and must leave at least one behind; taking every prompt is a rename.
  • Dragging works on the Prompts grid grouped by Topic: drag a row onto another group’s header, or onto the no-topic group to take it out of every topic. A selected row carries the whole selection with it. Dragging is a shortcut for a pointer; on a keyboard or a touch screen, use the prompt’s Move or the bulk bar’s Move…, which make the same change.
Merge into… and Split… are in the menu on a topic’s own page (Analytics → Topics, then the topic), offered when there is another topic to merge into and when the topic has at least two prompts. Every regroup (moving one prompt, a bulk move, a drag, a merge or a split) says the same thing before it happens: “Grouping is retroactive: topic numbers for past days will recompute. Prompts keep the engines they’re read on.” Both halves hold for all of them:
  • History recomputes. Topic numbers are worked out from the stored runs and where each prompt is filed now, so the topic a prompt lands in reports its whole history and the topic it left stops reporting it. Nothing needs rebuilding, and nothing waits for the next reading.
  • The engines do not change. A moved prompt keeps the engines it was read on, whichever topic it lands in, so a reorganisation never starts or stops reading a prompt on an engine. That is why a bulk move, a drag, a merge and a split still work when your credit balance is at zero; moving a single prompt with its own Move needs a balance above zero, so at zero, tick it and use the bulk bar’s Move… instead. The market is the exception: a prompt without a market of its own is read in the markets of the topic it lands in. A split’s new topic copies its source’s country but not a list of markets set on the source topic (its Markets menu item); set that on the new topic if the split prompts should stay in those markets.
None of the three is all-or-nothing, and each says what happened:
  • A merge that could not move every prompt keeps the source topic, so nothing you asked to move is left without a topic. It says how many moved and how many did not; merge again to finish.
  • A prompt added to the source topic while a merge is running is never deleted with it: it ends up with no topic.
  • A split that created its topic but could not move some prompts says so; those prompts stay in the original topic, and you move them from the prompt list.

​
Reading a topic rollup

Topic rates use the same shapes as everything else. Rates come back as {value, n} where n is the count of successful runs behind the number, and null means not measured rather than zero. Deltas come back as {value, real} and are suppressed to {real: false, value: null} when a two-proportion z-test at 95% cannot separate the change from noise. A topic’s n is the sum of its prompts’ successful runs, so topics cross the noise floor sooner than the prompts inside them. That is the practical payoff: a topic can prove a change in a week that a single prompt would take a month to demonstrate. Competitor rates roll up per topic too — and, as everywhere in Attensira, they share the same denominator as your own share of voice and do not sum to 100%. See Competitors for why.