Skip to content
SilktideHelp

Traits

Being mentioned tells you that assistants know you exist. Traits tell you what they think you are.

Every time an assistant answers a question, it says things about the brands it names: that one is expensive, that another is built for enterprise, that a third has patchy support. Those individual statements are . On their own each is a single sentence in a single answer and proves nothing. Grouped, they are the most useful thing in Presence: they are your positioning as the workspace's most influential intermediary describes it, whether or not it matches the positioning you paid an agency for.

A is one of those recurring qualities - "opaque pricing", "strong accessibility expertise", "built for developer workflows" - with every claim that fits it gathered underneath.

Why this exists

Three reasons, in rising order of usefulness.

You cannot fix what you cannot name. "We are criticized in 22% of answers" is a number you can worry about but not act on. "Assistants describe our pricing as opaque in 38% of the answers that mention us" is a sentence with a fix attached to it.

Positioning is contested, not declared. You can compare your own traits with your competitors' on the same grid. Sometimes the finding is that a rival owns a trait you thought was yours. Sometimes it is that a strength you genuinely have is simply never mentioned - you have it, and the workspace has not heard.

Criticism has an address. When an assistant says something unfavorable, it usually got it from somewhere. Silktide records the page cited alongside the claim, so an unflattering trait can be traced to the review, forum thread, or comparison article feeding it. Drill into any trait cell to read the claims and cited pages.

Where traits come from

Mostly, they find themselves.

  1. When Silktide reads a stored answer, it extracts the individual claims made about each brand you track.
  2. Claims that match an existing trait are filed under it, by name and known aliases.
  3. Claims that match nothing accumulate. When there are enough to see a pattern, Silktide clusters them into a new trait, which goes live immediately.

Traits are frozen once discovered, which is what makes them trendable. If the set were re-clustered every run, the same trait would appear under a slightly different name each month and no line on a chart would mean anything.

Each trait belongs to one arena - a workspace, or a brand for reputation topics. Two workspaces can both have "opaque pricing" and they are separate rows with separate claims, because the pricing conversation in one workspace is not the pricing conversation in another.

What a claim records

Each claim is one thing an assistant said about one brand you track:

TextThe statement itself, in short form
KindWhat it is (plain positioning), Strength, or Limitation
CriticismWhether the answer framed it as a complaint. Deliberately binary - the one judgment that changes how you should act - and overridable per trait on the manage screen, for the workspace where "expensive" is a boast
Compared withThe rival it was stated against, when the answer made a direct comparison
Cited pageThe source the assistant pointed to alongside the claim, when there was one
ResponseThe answer it came from, so you can always read the original words

Roughly half of all claims are plain positioning rather than praise or criticism, which is worth knowing before you read the grid as a scoreboard. "Aimed at enterprise buyers" is not good or bad; it is who assistants think you are for.

Reading the grid

On the Overview tab, the Traits report is a grid: traits down the side, brands across the top, criticisms in their own band at the bottom.

Each cell is that brand's share of its own responses that discussed the trait - not its share of the row. That is the important detail, and it is what makes two cells on one row directly comparable even when one brand is mentioned ten times as often as the other. A brand named in 20 answers with 5 mentioning its pricing reads 25%, exactly as a brand named in 200 answers with 50 mentioning pricing does.

An asterisk marks the clear leader on a row - highest, and beyond the margin of error, so a one-claim lead is not dressed up as dominance.

Three cell states are deliberately different:

  • A shaded percentage - the trait came up for that brand at that rate.
  • Discussed, never said - the brand was analyzed and this trait never attached to it. A real result.
  • A hatched cell - the trait never came up for that brand, or there were too few responses to say anything. An absence of data, not a zero.

The Profile tab draws the same data as one brand's ranked traits instead - any tracked brand's, your own by default.

Clicking any cell opens the claims behind it, so a percentage is always two clicks from the sentences that produced it.

Keeping the list useful

The taxonomy is capped, deliberately. A grid with a hundred rows is not a report; it is a spreadsheet nobody opens. You have four tools:

  • Merge two traits that are the same trait under different names. Claims move to the survivor.
  • Hide one that is noise, or too generic to act on. Hidden traits keep matching claims quietly, so the same trait is never re-proposed, but they stay out of reports.
  • Show again anything you hid too eagerly.
  • Flag or unflag a criticism, for the cases the model cannot judge - "expensive" is a complaint in most workspaces and a boast for a luxury brand.

When the arena is at capacity, Silktide will tell you to merge or hide something before it finds another. That is a feature: a short list is what keeps these reports readable.

Last updated

Was this page helpful?