AI search API use cases, step by step.Two customer journeys, from the first prompt to a weekly report.

For teams building an AI visibility product and agencies that report to customers every week. You own the pipeline: answers land in your storage on a schedule, and your own page shows the trend. New to the idea? Read what AI visibility tracking is.

1. 1

   Create a brand kit

   The brand kit describes the brand you track. Every prompt and run refers back to it, so set it up first.

2. 2

   Build a set of prompts

   Use the Prompt creation library to generate the questions your buyers ask. Tag each prompt and store it, because tags decide what runs together later.

   • Generate from the Prompt creation library
   • Tag by topic, persona, intent and market
   • Store the set

   # 2. Generate prompts for a brand kit, then store them with tags
   POST /v1/prompts/generate
   { "brand_kit": "northwind", "topic": "client reporting",
     "persona": "agency owner", "count": 25 }

   Illustrative sample.

3. 3

   Test a few prompts in Live mode

   Run a handful of prompts live and read the real answers before you schedule anything.

4. 4

   Review the output with a skill

   Open the results in Claude or ChatGPT and let a skill review them, so you see which prompts return useful answers and which do not.

5. 5

   Iterate until the prompts are right

   Use what the review found to tune the set. Repeat steps 3 to 5 until the answers match what your buyers ask.

   • Delete prompts
   • Update prompts
   • Create prompts
6. 6

   Create a destination for the output

   Point the output at storage you own.

   • S3 bucket
   • BigQuery

   # 6. Register where runs are written
   POST /v1/destinations
   { "type": "bigquery", "dataset": "ai_visibility" }
   # → { "id": "dst_bq_01" }

   Illustrative sample.

7. 7

   Schedule runs by tag in Claude

   Create a scheduled task in Claude that runs specific prompts by tag on a regular basis. Pass the destination with every run.

   • Weekly prompts: standard mode, results land 10 min to 1 hr later, destination passed
   • Daily prompts: destination passed as a parameter

   # 7. One call per cadence, e.g. run_prompts_by_tag(tags, destination)
   # Weekly, standard mode (delayed 10 min to 1 hr)
   run_prompts_by_tag(tags=["weekly"], destination="dst_s3_01")
   
   # Daily
   run_prompts_by_tag(tags=["daily"], destination="dst_bq_01")

   Illustrative sample.

8. 8

   Check the output

   Open the S3 bucket or BigQuery table and confirm each scheduled run landed with its tags.

9. 9

   Show the analytics in your app, Claude Artifacts, slide decks, billboards… anywhere

   Run a skill that reads the stored answers and puts the analytics wherever they are needed: your app, a Claude Artifact, a slide deck, even a billboard. Keep it current for your product or the client's weekly report.

For teams that want reporting without building a pipeline. Answers stay in MoltMentions, which runs the schedule, sends notifications and turns the answers into AI share of voice analytics and recommendations.

Steps 1 to 5:
the same as Scenario 1

1. 1. Create a brand kit
2. 2. Build a set of prompts
3. 3. Test a few prompts in Live mode
4. 4. Review the output with a skill
5. 5. Iterate until the prompts are right

6. 6

   Use MoltMentions as the destination

   Skip the bucket. Runs are stored in MoltMentions, where the pre-built AI Visibility Analytics and Recommendations read them.

7. 7

   Create a schedule in MoltMentions

   Set which prompts run and how often inside MoltMentions. No scheduled task to maintain.

8. 8

   Connect Slack for notifications

   Send notifications to a Slack channel, so the team hears about changes without opening a dashboard.

9. 9

   Manage it all through MCP

   Use the MoltMentions MCP server to manage visibility, reporting and recommendations from your AI assistant.

   Copy

   # Claude Code. Then run /mcp in a session to sign in.
   claude mcp add --transport http --scope user Thrivestack \
     "https://api.app.thrivestack.ai/mcp"

   Illustrative sample.