Skip to content
SilktideHelp

Answers screen

Every AI figure in Presence is derived from stored answers. This is where those answers live.

A response is one stored record of what an AI assistant answered for one prompt at one moment. When a number looks wrong, surprising, or too good, this is the screen that settles it. Search results pages are stored the same way but live on their own screen, Results, because a results page between two AI answers reads as neither.

The list

Open AI → Answers in the sidebar. One row per measured answer.

Each row carries the prompt, the assistant, when it ran, and what was found in it. The list inherits the filter strip's scope, so arriving from a topic, a prompt, or a brand narrows it to the answers behind that view.

Tested is when that run was asked for, not when that particular assistant got back to us. Every answer in a run carries the same time, so a batch stays together in the list however long individual assistants took, and a run that started just before midnight counts as one day rather than being split across two. Assistants appear in a consistent order within each run, so the list reads the same way every day.

One column reads the list from your side: portrayal says whether that answer praised your brand, criticized it, did both, or named it and passed no comment. It is blank when the answer passed no judgment at all - usually because it never named you, occasionally because it only name-dropped you in a footnote, which carries no opinion and is left out of your sentiment figures for the same reason. Narrow the brand filter to a single competitor and the column becomes theirs, so you can scan a rival's coverage the same way. The four states are the ones the Reputation screen counts. See Sentiment for what they mean.

Status is not shown while the list is filtered to one status, which it is by default - a column reading "Responded" beside every row tells you nothing you did not just ask for. Set the status filter back to All and the column comes back.

The Arena filter narrows the list by what kind of question was asked: questions about the workspace, questions about a brand itself, or both together (the default). It also offers Accuracy - the answers behind the Accuracy screen, which are collected the same way but kept out of every other view, because an accuracy question names its brand on purpose and would flatter every visibility figure it touched. Choosing Accuracy is the way to read one of those answers in full, highlights and all.

You can search, sort, and export.

Reading one response

Opening a row shows the stored record: the AI's answer in its own words. That is the source, not a summary Silktide wrote.

Named brands are marked in that text. Praise and criticism are highlighted inline, so you can see who was named and how they were treated without leaving the words. Hover a wash for the short explanation.

The sidebar lists who was mentioned and how positively or negatively they were regarded. Click a brand to highlight the sentence or result it came from.

Everything else Silktide read out of the same record sits in that sidebar too:

  • History for the same prompt on the same assistant. Earlier answers appear in a rail you can show or hide; pick any row to open that answer without losing your place in the list. The choice is remembered on this device.
  • Brands named in the response, in order, with the tone of each mention.
  • Brands also mentioned - names that appeared without being offered as an answer.
  • Searches the assistant ran before it answered, in order - its own research, not yours. Words it added that nobody asked for are its assumptions about the question, and a search aimed at one site is a site it already trusted. Only some assistants report this, so the panel is absent on most answers. See the Searches screen.
  • Sources, with the same locate-in-the-answer behavior, so you can see what each citation was attached to.
  • Ads the assistant served beside the answer, when the platform shows them. These are the assistant's ads, not Silktide's. See Advertising in AI.
  • Untracked names the assistant used that you do not follow yet.

From here you can add an untracked name as a competitor, merge one that is really an alias of a brand you already have, or jump to a report on any brand or topic mentioned.

Not every response is an answer

Three outcomes are recorded, and two of them are normal:

  • Answered - the assistant replied and Silktide analyzed it.
  • No answer shown - the assistant deliberately showed nothing. This is a real result, most common on Google AI Overviews and AI Mode, and it is counted as an absence rather than an error.
  • Failed - no answer could be obtained. Nothing is recorded, so it is left out of your figures entirely rather than counted against you.

Occasional failures are expected and need no action. A pattern of them - one assistant failing on most prompts, day after day - is worth telling support about.

Last updated

Was this page helpful?