Web research

Web research uses OpenAI to search public sources. One research can cover several competitors, markets, or custom topics, each with its own questions. When a question names an App Store app, Zentrik also reads its rating and recent reviews directly; if several apps match, it asks which one. No provider credentials are needed.

1. Open Discovery → Research and choose New research. Choose Competition, Market, or Public feedback and enter the subjects, or choose Custom research and ask a question.
2. Review the plan: the questions, source scope, and research window. If several App Store apps match, choose the one you mean; the plan shows sample reviews from it. Choose Refine to change anything or remove an app.
3. Choose Start research. Results stay in the research until you add them to Discovery.

Under Research options, choose Establish a baseline to include relevant older and undated documents, or Recent developments for a dated window of 1–90 days.

People with workspace edit access can run research. Owners and admins manage saved research and its update frequency. The questions and scope are sent to OpenAI.

Find saved research by topic, question, or product from the research home. In a report, Findings shows each question as one line; select it for the full answer and the reviews and pages it cites. Filter to Changed or Needs evidence. When research covers several subjects, choose one at the top of the report; its findings, ratings, and themes follow.

Sources groups what the run read, such as App Store reviews or web pages. Open a group to search and filter every saved item. If an answer cannot be prepared, the saved sources remain available.

The toolbar places every run on a timeline; choose one to read it. A later run of the same plan is compared with the run before it, and a changed finding says what changed under its answer. A changed plan starts a new comparison.

When subjects share questions, Compare places their answers side by side. From the research home, Compare research brings together any saved research with shared questions, without starting new research.

To correct a finding, use Refine this research below the findings, then Review revised plan and Run revised research. To change the whole plan, choose Review and run again.

Each run reads the US App Store rating of every confirmed app, including stars left without a written review. Store ratings sets the all-time rating beside the average of the ratings that arrived since the last run. The average appears once enough new ratings arrive to state it within a tenth of a star. To measure a longer period, such as since a release, choose an earlier run beside Store ratings.

When a run reads at least five new reviews of an app, What reviewers raise groups them into problems, requests, and what reviewers value, with how many reviews raise each and how that changed since the last run. Select one to read exactly those reviews. Reviews are the most recent ones from the US storefront: a sample, not every customer.

1. Choose Add to Discovery, pick the product the sources become evidence for, and add them. Each source becomes a Signal: a web page keeps its link, dates, and an AI summary; a review keeps its author’s words, rating, date, and link. Company announcements stay the company’s claims.
2. After the addition, choose Only when I ask, Weekly, or Daily under Updates, beside the report. Automatic research starts only after you save a frequency, and each update uses your workspace’s AI allowance.

Later runs look for new sources, with a three-day overlap. Pages and reviews already added are skipped; edits to a page already read are not detected. Review and add a revised plan’s sources before it replaces an existing schedule.

Choose Pause updates to stop automatic runs. Pause before archiving; archiving keeps the run history and the Signals already added. Research frequency is separate from Brief emails.

You can also read and run research from a connected MCP agent or the Research API.

Sources per section sets how many qualifying sources a section keeps. A first run makes one search round per section and keeps three by default; scheduled runs make up to three rounds to reach the target. Rejected and duplicate sources do not count, and a target is not a promise that matching evidence exists.

A run reads up to four confirmed App Store apps: 40 recent reviews in total, at most 25 per app, from the US storefront. A section limited to the App Store skips web search.

Every source must fall inside the research window. A review keeps its own date, and a month-only date qualifies when the whole month is inside the window. Recent research skips undated pages; a baseline can keep relevant ones, marked as undated.

Coverage notes name topics that stayed unverified, which is different from a topic researched with no new findings. When a section finds no accepted sources, or falls short with limited coverage, the next run keeps the same window. Search never guarantees complete coverage of a market.