Track Brand Mentions in AI Answers

When a client asks “How visible are we in ChatGPT this month?”, the old reporting playbook stops working. There’s no ranking position to look up and no Search Console export to attach, and the same prompt run twice rarely returns the same answer. Each AI response is generated on the fly, citations sit inside paragraphs rather than in a SERP block, and the prompt set that triggered the brand to appear is invisible from the outside. This use case walks through how a GEO consultant or brand agency can turn that into a report a client will trust, using our LLM Mentions API for the data.

The problem: AI answers don’t have rankings, and you can’t refresh them

Tracking brand visibility in LLMs is harder than tracking it in classic SERPs for one reason: the data isn’t sitting in a public index that you can poll. Manually prompting ChatGPT or Gemini and screenshotting the result is the most common first attempt, and it stops scaling almost immediately. In practice, a single prompt rarely returns the same answer twice. The prompts that drive real AI traffic for a brand are not the ones a marketer would guess at on a Monday morning. A hand-collected screenshot library can’t be filtered, aggregated, or compared week over week either.

Agencies need a report they can defend in a client meeting: one that says “your brand was mentioned in X answers, across Y prompts, alongside competitors A, B, and C, with this much AI search volume behind it.” Producing that by hand isn’t realistic at any meaningful client volume. Our LLM Mentions API handles the collection step, so the GEO workflow can focus on interpretation and recommendations.

What tracking brand mentions means in GEO

Before pulling any data, it helps to split the question “How visible is the brand in AI answers?” into three smaller ones. Each one maps cleanly to a different endpoint in the LLM Mentions API:

• Where is my brand showing up, and in what context? Answered by the Search Mentions endpoint, which returns the prompts, answers, and cited sources that include the brand.
• How does my visibility compare to competitors, side by side? Answered by the Cross Aggregated Metrics endpoint, which compares multiple brands or domains in a single request.
• Which domains and pages are AI models treating as authoritative in my category? Answered by the Top Domains and Top Pages endpoints, which surface the sources most frequently cited for a given topic.

This three-question framing keeps a weekly GEO report tight: one section per question, one endpoint per section. The fiddly part is defining the targets correctly. The search_scope, match_type, and other parameters all change what you get back. Our dedicated Help Center article covers the parameter combinations in detail, and it’s worth keeping open while you set up your first project.

The use case: a weekly brand mention monitor for one client

Here’s a workflow you can run for a single client, every Monday, in under five minutes of compute time once the prompts and competitor list are defined.

Step 1. Decide on the brand and the two to four competitors that will anchor the share-of-voice comparison. For a design-tool brand, that might be Figma, Adobe, Sketch, and Canva. For a CRM brand, HubSpot, Salesforce, Pipedrive, and Zoho.

Step 2. Pull all AI answers that mention the brand. The request below uses the Search Mentions endpoint to return ChatGPT responses that contain “Figma” in the response text, sorted by AI search volume so the most-trafficked prompts come first. The search_scope is set to answer so the brand has to appear in what the model said back to the user, rather than only in the user’s prompt.

Request example:

[
  {
    "target": [
      {
        "keyword": "Figma",
        "match_type": "word_match",
        "search_scope": ["answer"]
      }
    ],
    "platform": "chat_gpt",
    "location_name": "United States",
    "language_code": "en",
    "order_by": ["ai_search_volume,desc"],
    "limit": 100
  }
]

Endpoint used: api.dataforseo.com/v3/ai_optimization/llm_mentions/search/live. See the Search Mentions docs for the full parameter list.

Response example (single item, trimmed answer text):

{
  "version": "0.1.20251208",
  "status_code": 20000,
  "status_message": "Ok.",
  "time": "1.4823 sec.",
  "cost": 0.001,
  "tasks_count": 1,
  "tasks_error": 0,
  "tasks": [
    {
      "id": "06251242-1535-0634-0000-9d825a2445b0",
      "status_code": 20000,
      "status_message": "Ok.",
      "result": [
        {
          "platform": "chat_gpt",
          "model_name": "gpt-5-5",
          "location_code": 2840,
          "language_code": "en",
          "question": "why are people ditching adobe?",
          "answer": "A lot of people are not ditching Adobe ... Figma for UI/UX design ... Affinity Photo, Affinity Designer ...",
          "sources": [
            {
              "source_name": "Fstoppers",
              "title": "Why Photographers Are Leaving Adobe in 2026",
              "domain": "fstoppers.com",
              "url": "https://fstoppers.com/software/real-reason-photographers-are-leaving-adobe-901527"
            },
            {
              "source_name": "MakeUseOf",
              "title": "Why I Wouldn't Recommend Anyone Get Adobe Creative Cloud in 2024",
              "domain": "www.makeuseof.com",
              "url": "https://www.makeuseof.com/why-dont-recommend-getting-adobe-creative-cloud/"
            }
          ],
          "ai_search_volume": 4957,
          "monthly_searches": {
            "2026-05": 4957,
            "2026-04": 5568,
            "2026-03": 5132
          },
          "first_response_at": "2025-10-25 12:14:49 +00:00",
          "last_response_at": "2026-06-02 17:07:36 +00:00",
          "fan_out_queries": [
            "why are people leaving Adobe alternatives Creative Cloud complaints 2025"
          ]
        }
      ]
    }
  ]
}

That one item alone is useful. Figma appears inside a ChatGPT answer to the prompt “why are people ditching adobe?”, which had an estimated 4,957 AI searches in May 2026. The brand was probably not aware it was surfacing in that competitive context at all. The sources array names the four publications ChatGPT drew from to write the answer, which gives the GEO team a starting list of domains to pitch, partner with, or write about.

Step 3. Run a head-to-head share-of-voice analysis across the competitor set. Use the Cross Aggregated Metrics endpoint to compare all four brands in one request. You get a single task back, with one row per brand and the same metric set across all of them. We’ve covered the parameter shape and read-back logic in a dedicated Help Center article if you need a deeper walkthrough.

Step 4. Log the run. Store the result in a Google Sheet, BigQuery, or whatever the agency already uses for client reporting, with the run date as a key. The next week’s pull gives you week-over-week deltas without any manual prompting.

Reading the data: what to report, what to ignore

The Search Mentions response contains roughly twenty fields per item. For a weekly client deliverable, four of them carry most of the signal:

• ai_search_volume: the estimated number of times a prompt was sent to the model in the most recent month. This is the metric that turns a single mention from “we appeared once” into “we appeared in answers to a prompt with ~5,000 monthly AI searches.” Without it, you can’t prioritize.
• question: the user prompt that triggered the brand mention. This is the closest thing GEO has to a keyword report, and it should drive content planning. If the brand is showing up on competitor-comparison prompts but absent from “best [category] in 2026” prompts, that’s a content gap to brief.
• sources: the array of URLs the model cited when constructing the answer. Treat this as your authoritative sources list for the topic. If the same five domains dominate the sources across most mentions in your industry, those are the publications worth pitching, partnering with, or guest-posting on.
• monthly_searches: the prompt’s AI search volume over time, month by month. Use this to call out trending prompts in the report instead of treating every mention as equally important.

Fields like model_name, fan_out_queries, and brand_entities are useful for deeper analysis. The model_name lets you split the report by ChatGPT model version when comparing data across longer time periods. The fan_out_queries array surfaces the sub-queries the model ran in the background to build its answer. For most workflows, the four fields above are enough for a weekly executive-facing slide.

Conclusion

You can’t rank AI answers the way you ranked SERPs, but you can measure them. The data you need is mention data, pulled from real model responses on a schedule instead of collected by hand. The workflow above is intentionally small: one brand, one competitor set, three endpoints, four reportable fields. Realistically, the same agency can scale this to a roster of fifty clients without rebuilding the pipeline itself; only the prompts, the competitors, and the storage layer change. Our LLM Mentions API docs have the full parameter reference, and our blog post on data-driven AI visibility insights covers the broader GEO context. To start pulling brand mention data for your own clients, try for free. No commitment, pay only for the data you use.