Attest

View saved version →

---
name: compare-audiences
description: Compare how different groups answered an Attest Measure question, such as age groups, saved audience segments, or markets like the UK and US.
---

Use this when the user wants results split by group: two or more audiences, saved segments, demographic bands, or markets.

1. Find the study and read its structure to find the question and the surveys it ran in.
2. Work out how each group is defined:
   - A group the user names may be one of their saved segments. Check those first and use a matching segment's own filter as it is.
   - Build a filter yourself only when no saved segment matches, using the fields and values the connection reports as valid for that market.
   - Markets are usually separate surveys rather than filters. Compare them by reading each market's survey.
3. Read results once per group, one survey at a time, each with that group's filter. Never read the whole sample once and estimate the groups from it. If two groups come back with identical figures, check that each request carried its own group's filter before reporting. Identical results can be genuine, but they're also how a dropped filter shows up.
4. When comparing markets, compare like with like. A question or option hidden from one market means it wasn't shown there, not that nobody chose it.
5. Present the groups side by side for each answer, with every group's base, and point out the biggest differences. Treat any group below the reliable base the results flag as indicative only, and say when a difference is too small to read much into.

If the user names a group you can't define from their segments or the available fields, say so rather than substituting a near match.

View saved version →

---
name: survey-results
description: Answer general questions about the user's Attest Measure studies by finding the study and reading its published survey results, boards and key findings. For change over time or splitting results by group, the tracker-trends and compare-audiences skills go further.
---

Use this when the user asks about their Attest research: a study, a survey, a tracker, a wave, or how an audience answered a question.

1. Find the study the user means by searching their studies. If several match, ask which one rather than picking.
2. Read the study's structure before any results. It tells you the surveys (one per wave and audience), the questions and their answer options, and the IDs the other tools need.
3. Read results one survey at a time. To compare waves, markets or audiences, request each survey separately and compare the answers yourself.
4. To slice by a group of respondents, prefer a segment the user has already saved. Build your own filter only when no saved segment matches, using the fields and values the connection reports as valid.
5. When the user asks what a study found, check its boards and key findings before working it out from raw results.

Report the numbers the tools return, with the base (how many people answered) alongside percentages, and say when a base is small. Name waves by their published date.

If the user asks for something the connection's tools can't do, say so plainly and point them to the Attest dashboard rather than attempting it.

View saved version →

---
name: tracker-trends
description: Show how an Attest Measure tracker's results have changed across waves, such as whether awareness or consideration moved since the last wave or over a date range.
---

Use this when the user asks how a result has changed over time: since the last wave, across a date range, or over the life of a tracker.

1. Find the study, then read its structure to list its waves. Each wave has a published date and its own surveys. Work out which waves the question covers: the latest and the one before it for "since last time", or every wave whose published date falls in the range the user gave.
2. Confirm the question exists in every wave you compare and that its answer options match. If a question was added later or its options changed, compare only the waves where it is the same, and say so.
3. Read the results one survey at a time for the same question in each wave. If a wave has several audiences or markets, keep each audience in its own line rather than pooling them, unless the user asked for a total.
4. Present the waves in date order, named by published date, with each wave's figure and base side by side, and the change between them. Give changes in percentages as percentage points. Keep other measures, such as average rank or a preference score, in the units the results use.
5. Describe movement plainly. A change of a few points between modest bases may be noise, so call it a small change rather than a trend. Treat any wave below the reliable base the results flag as indicative only.

Keep the answer about the movement the user asked for. Don't add a figure that blends several waves into one number.