From evidence to decisions: Research workspace

Think of the Research workspace as a private notebook beside your AI checks. The check shows what assistants said. The notebook helps you decide what to do about it and keep the evidence for what happened next.

Open Research workspace in the AI visibility menu, or Question research in the brand workspace followed by Open your Research workspace. Choose the business and market. Research items are private to your account and are not included in shared checks or editor invitations.

On a phone, open the dashboard menu, choose AI visibility, then Research workspace. The six section buttons switch between the same tools available on a larger screen. In Human review, swipe the summary table sideways if all its columns do not fit.

A simple example

Imagine you run a bakery and customers keep asking whether you make gluten-free cakes.

1. 1Save that question and where you heard it under Customer questions. If it is only your idea, label it as a hypothesis.
2. 2Use a check to see how AI assistants describe your bakery. In Human review, compare a retained answer with the evidence and record your judgment.
3. 3Under Track an intervention, record a plan to clarify the cake information on your website. Pick a baseline check from before the change and some separate questions you intend to leave unaffected.
4. 4After making the change, collect a later check with matching questions and settings and attach it as the follow-up. See what moved in both groups. Add suitable business totals if you have them.
5. 5If you want to explain the result to someone, prepare a private Case studies draft with the evidence and its limits.

This can help you decide what to improve next. It cannot promise that assistants will change their answers or that a change in sales was caused by your website edit.

What you supply and what it costs

You supply the decision, customer-question sources, human judgments, planned changes and any real survey or interview results. The workspace does not recruit people or run a survey for you, and AI answers are not human respondents. It organizes your evidence and calculates comparisons where the records are suitable.

Saving evidence does not use check credits. Running a new check from selected questions uses the usual check credits or payment, shown before you run it. Research you arrange with people or another provider is separate.

Decision brief and customer questions

Use Decision brief to record the decision, audience, coverage and gaps. In Customer questions, save the exact question and its origin, dated evidence, audience and stage. Keep untested ideas labelled as hypotheses. Select 3 to 12 questions, then choose Use selected questions to review a new check before running it. The usual check credits apply.

Human review

Choose a check, assistant and question. Inspect the retained excerpt and open the full check for the complete answer and sources. Review brand identification, tone, local relevance or citation support. Record the judgment, reasoning, reviewer and dated source. Uncertain judgments are excluded from the error-share denominator. Each answer and criterion counts once, with the newest judgment used. These are your team's judgments, not an independent certification.

Track an intervention

Choose a baseline check collected before the intervention day. Record the change, hypothesis and other possible influences. Choose separate affected and unchanged question groups. Save the plan before acting. Add a follow-up when it is available. The comparison requires recorded, matching collection settings and questions. Old checks without recorded settings remain exploratory.

The result compares only cells answered in both checks. It subtracts the unchanged group's movement from the affected group's movement, but does not prove causation. Optional business outcomes use owner-supplied counts, a source, equal non-overlapping periods and explicit attribution limits. Use Link existing Analytics evidence to compare dated website records for the same property and definitions. The comparison covers the whole property, not just the check market. Only references are saved, and disconnecting Google removes its data from the comparison. These linked figures are not copied into case-study exports. Do not paste connected Google figures into research notes.

Human research

Record real surveys, interviews, concept tests, pricing studies or licensed market datasets. Explain recruitment, sample selection, population, weighting and limitations. Paste tab-separated aggregate rows with these headers: question, answer, segment, count, denominator. Each question and segment needs a consistent denominator and mutually exclusive answer categories. Up to 200 rows can be saved. Do not include personal respondent details or generated responses.

Private evidence export

Export private research evidence downloads the saved notes, references and review declarations as JSON. It does not copy linked Analytics figures.

Case studies

Choose an intervention and any saved studies, then record your interpretation, limits, consent and review references. Export evidence draft downloads a private Markdown document. It does not publish the case study or certify an independent review. Missing or deleted evidence is shown as unavailable.

Repeatability

New checks record collection settings and repeat up to two additional questions with up to three selected assistants, alongside the category question's repeats. Open Repeatability and comparison settings on the check to see observation counts and prominence ranges. Repeats do not enter the headline score. Small repeat samples describe observed variability, not a population margin of error. The PDF, PowerPoint and Markdown exports carry the same explanation.