DataSag guides · Updated 27 October 2026

Build an AI visibility prompt test you can repeat

To evaluate AI visibility in DataSag, save an exact customer question, run it with your connected OpenAI API key, and inspect the returned answer and cited URLs. Repeat the same question with the same model and target to build comparable observations. Each run uses provider credits and represents one API response with web search.

Choose a question that represents a decision

Begin with a customer problem your product can actually solve. For website analytics, a question might ask how to install tracking on a specific framework or how to compare setup requirements. Keep the first observation focused enough that you can explain why a citation would be useful.

Separate discovery questions from brand questions. Asking which tools support a workflow measures a different situation from asking whether DataSag supports that workflow. A question that names your brand is useful for checking product accuracy, but it should not count as independent category discovery.

A large prompt library is an inventory of hypotheses. It is not evidence that those questions have search demand, that people use the exact wording, or that a website appears in the answers. Review questions against support conversations, product workflows, and actual search reporting before spending credits on them.

Write the test record before the run

Give each question a stable identifier in your own evaluation notes. Record its intent, audience, target website, exact wording, and why you chose it. Include an expected evidence type, such as a source link to a setup guide or a correct description of the installation step.

Keep brand inclusion explicit. If you use a prompt that names a competitor or your product, label the prompt accordingly. Do not silently rewrite a weak result into a brand-led question and treat the new answer as an improvement on the original test.

Choose a small initial set you can inspect by hand. A thousand unreviewed answers can leave you with less useful evidence than a dozen samples whose citations you actually open. Expand the set after you understand the common errors and can keep the review consistent.

Run a selected prompt in DataSag

Open AI visibility in the website project and choose AI prompt tests. Connect your own OpenAI API key using the provider connection form. Save a customer question, select it from the prompt list, and review the provider-credit confirmation before you run the test.

  1. Save the exact question you intend to repeat.
  2. Select that question as the prompt to run.
  3. Confirm that the provider may use your credits for the run.
  4. Run the selected prompt and wait for the saved report.
  5. Read the answer and open its citations before marking the observation as useful.

Review citations and product accuracy separately

DataSag extracts URL citations from the returned answer and marks domain matches. Inspect the source address and title. Confirm that the source page exists and supports the sentence it accompanies. A matching domain alone does not establish that the assistant described your product correctly.

Use a separate note for content accuracy. For example, an answer may cite your guide but describe a capability your product does not provide. That is an editorial problem worth investigating even though the domain-cited field is positive.

A result with no returned citations is inconclusive about wider visibility. It records an absence in that response. A request that fails or does not complete should also stay outside successful-sample counts. Repeating a failed provider call is different from observing a successful answer that omits you.

Compare equivalent observations

DataSag compares saved reports with the same prompt, model, and target. Its interface uses up to the 40 most recent project reports when showing comparable samples. Read the sample count alongside the number of samples citing the domain.

When the model, prompt, or website changes, start a separate comparison. Keep a record of content release dates and the affected URLs. Repeated responses can vary even without a release, so one before-and-after pair is a starting observation rather than a stable visibility trend.

The provider is the OpenAI API with web search. It does not reproduce the ChatGPT consumer app or measure every assistant. Do not describe its sample proportion as market share, overall AI ranking, or the percentage of all users who see your brand.

Use the result to choose an action

If an accurate competitor guide is repeatedly cited for a task you support, inspect the missing answer on your own website. Add the actual setup requirements, code, constraints, and verification evidence that would help a customer.

If your page is cited but the next step is unclear, improve the landing page. If citations exist but referral visits remain sparse, keep the two observations separate and investigate intent. The evaluation should lead to a specific content or product decision, not a larger number of automated requests.

Sources

Continue with a practical guide

Install analytics on your platform or explore DataSag search and AI visibility.