Key takeaways
- ChatGPT has no official ranking API or Search Console equivalent, so a Looker Studio dashboard needs a data feed: a visibility tracker's export, or your own prompt-run log in Google Sheets or BigQuery.
- Track a fixed prompt set segmented by prompt type (organic, brand-specific, competitor comparison). One prompt tells you almost nothing.
- The core metrics are visibility rate, citation rate, position in the answer, share of voice and sentiment, plus AI referral traffic from your analytics.
- Looker Studio does the visual part well. The hard part is deciding what goes in the data source, so spend most of your time there.
- Use trend lines and monthly views, not daily snapshots. AI answers are noisy.
Why a ChatGPT ranking dashboard is different from an SEO dashboard
In Google you have a rank. Position 4 for a keyword is position 4, more or less. ChatGPT does not work like that. The same prompt can return different brands, different sources and different ordering from one run to the next. There is no "ChatGPT Search Console" to pull from.
So a "ranking" dashboard for ChatGPT is really a visibility dashboard. You are measuring how often your brand appears across a set of prompts, where in the answer, which of your pages get cited, and whether any of it sends traffic.
That changes how you build it. In a normal SEO dashboard you connect Search Console and you are done. Here, you have to create the data first.
One thing worth knowing before you design anything: ChatGPT answers are built from several web searches, not one. Promptwatch's query fanout data shows average fanouts per response at about 2.15 in early December 2025 and roughly 1.84 by early March 2026, with average query length falling from about 117 characters to the high 80s. Behaviour shifts after model updates, which is a good reason to keep a time series rather than a single report.
Step 1: Decide what the dashboard must answer
The SE Ranking Looker Studio guide starts with the same advice that applies here: analyze your data and decide what to display before you touch the tool. For a ChatGPT dashboard, I would write down four questions:
- Are we mentioned more or less often than last month?
- Against which competitors do we win or lose?
- Which of our pages get cited, and which prompts trigger them?
- Does any of this produce visits and conversions?
If a chart does not help answer one of those, leave it out. Dashboards die from clutter, not from missing widgets.
Step 2: Choose your metrics
Here is the set I would start with. Resist adding more until the team actually uses these.
| Metric | What it tells you | Typical problem it reveals |
|---|---|---|
| Visibility rate | Share of tracked prompts where your brand appears | Coverage gap |
| Position in answer | Where you are first named or listed | Prominence gap |
| Citation rate | Share of responses citing your own domain | Content or crawlability issue |
| Share of voice | Your mentions vs. all tracked brands | Competitive gap |
| Sentiment | How the answer describes you | Reputation issue |
| Citations per response | Benchmark line for how many sources ChatGPT uses | Context for your own numbers |
| AI referral sessions | Real visits from chatgpt.com and others | Business impact |
Read them together. Low share of voice with decent position is a coverage problem: you need to show up for more prompts. Mentions without self-citations usually mean the model knows about you but is pulling facts from someone else's page.
A note on benchmarks
Be careful when you add benchmark lines. Promptwatch's average sources per response data (published September 2025, ChatGPT messages where web search triggered) puts ChatGPT at about 5 cited sources per answer, with AI Overviews near 10 and Perplexity almost exactly 10. A newer report on unique domains per ChatGPT response (Sep 28 to Oct 4, 2026, responses with at least one citation) found 47.6% of responses cite 10 or more domains and 66.1% cite at least 8.
These look contradictory and they are not directly comparable. They cover different periods and different counting methods, and the newer one counts unique domains per response. Pick one, label its date on the chart, and do not put both on the same axis.
The same report splits by prompt type, which is useful for your segmentation: 52.5% of organic prompts cite 10+ domains, versus only 21.3% of brand-specific prompts. If you mix prompt types in one average, your numbers will move for reasons that have nothing to do with your performance.
Step 3: Get the data
This is where most tutorials skip ahead. You have three realistic options.
Option A: Export from an AI visibility platform
The least painful route. A platform runs your prompts daily against the real ChatGPT interface and gives you visibility, position, citations and sentiment. Then you push that to Looker Studio through a native connector, CSV or Sheets export, or an API.
Promptwatch is the one I would look at first here, since its Professional plan and above include a Looker Studio connector, plus CSV and Excel exports of prompts, responses and citations, and a REST API. It also tracks the real ChatGPT UI rather than just API output, which matters because user-facing answers can differ from API answers. Check promptwatch.com/pricing for current plan limits.
Other trackers in the catalog that you could evaluate for export options include Otterly.AI, Peec AI and LLMrefs. Confirm connector or export support on each vendor's site before you commit, because I have not verified it for all of them.
For more options, the AI rank tracker directory at ai-rank-tools.com lists a lot of them.
Option B: Your own prompt log in Google Sheets
If you want to start free, you can run a fixed set of 30 to 50 prompts by hand once a week and log the results. It is tedious but it works, and it teaches you what to measure before you spend money.
Use one row per prompt per run, with these columns:
date | prompt | prompt_type | model | brand_mentioned | position | own_domain_cited | cited_domains | sentiment | competitor_1_mentioned | competitor_2_mentioned
Keep prompt_type to three values: organic, brand, comparison. Keep brand_mentioned as 1 or 0 so Looker Studio can average it into a rate.
Be honest about the limits. Manual runs are small samples, they are hard to repeat consistently, and you will stop doing them around week six. Good for a pilot, not for reporting to a CFO.
Option C: BigQuery for larger setups
If you already have a data warehouse, load tracker exports or API responses into BigQuery on a schedule and connect it as a data source. This is the right choice when you want to join AI visibility with CRM or revenue data. It is overkill for a single brand with 50 prompts.
Add real traffic data
Visibility is not revenue. Pull AI referral sessions from GA4 into the same report by filtering source or medium for chatgpt.com, perplexity.ai, gemini.google.com and similar. Google Search Console data is worth blending too, so you can find pages that rank in Google but never get cited by AI.
Step 4: Connect the data to Looker Studio
The Looker Studio home screen has three areas that matter: Reports, Data sources and Explorer. The Supermetrics tutorial describes the Create menu offering a blank report or a new data source, with a gallery of connectors (it says 1,000+, though that post is from 2024, so expect the number and the product naming, Looker Studio vs. Data Studio, to have moved).
The steps:
- Go to Looker Studio and click Create, then Data source.
- Pick the connector: Google Sheets, BigQuery, a platform's native connector, or a CSV upload.
- Authorize and select the sheet or table.
- Check the field types. Dates must be Date, rates must be Number, and prompt_type should be Text.
- Click Create report and add the data source.
If you want a general walkthrough of scorecards, line charts and tables before you start, the Coupler.io tutorial covers them step by step.
Fix the fields before you chart
Two calculated fields save a lot of pain.
Visibility rate:
AVG(brand_mentioned)
Citation rate (own domain cited in responses):
AVG(own_domain_cited)
Because the source columns are 1 or 0, an average gives you a percentage. Format it as Percent in the field settings.
Step 5: Lay out the dashboard
I would build one page first, then split into more only when someone asks for it. Here is a layout that works.
Top row: scorecards
Four scorecards: visibility rate, average position, citation rate, AI sessions. Turn on comparison to the previous period so each shows a delta. A scorecard without a comparison is decoration.
Second row: trend
A time series of visibility rate with weekly granularity. Add a second line for your top competitor if your data has it. Use weekly or monthly buckets, not daily. Daily data from AI models is jumpy, and people will panic over noise.
Promptwatch's sources-per-response report makes this point about Copilot: it swung from under 2 citations to nearly 17 within a few weeks, then settled low, and the report recommends judging it on monthly trends. The same logic applies to any model you track.
Third row: segmentation
A bar chart of visibility rate by prompt_type. This is the chart that stops you from fooling yourself. If your overall number is climbing only because you added easy brand prompts, this chart shows it immediately.
Fourth row: tables
- Top cited pages from your own domain, sorted by citation count
- Prompts where you are absent but a competitor is present (your gap list)
- Most frequent competitors by share of voice
The gap list is the part people will actually act on, so give it room.
Filters
Add drop-downs for model, country, prompt_type and date range. Put them at the top, and keep prompt_type visible by default.
Step 6: Blend visibility with traffic
In Looker Studio, blending joins two data sources on a shared key. The usual keys here are landing page URL or date.
A useful blend: join your cited-pages table with GA4 sessions from AI sources on page URL. Now you can see which cited pages actually drive visits. Another: join with Search Console on page to flag pages that earn Google clicks but zero AI citations.
A few warnings. Blends in Looker Studio are limited and can get slow. Normalize URLs first (trailing slashes, http vs. https, query strings) or the join will drop rows silently. If you are doing heavy joins, push them into BigQuery instead.
Step 7: Add the crawler layer if you can
If you want to understand why you are or are not cited, add AI crawler data: which bots hit which pages and whether they get errors. This needs server or CDN logs, and some platforms provide it pre-packaged. A page that AI crawlers cannot fetch will not get cited no matter how good the content is.
For tools that can help here, see the GEO software directory at bestgeosoftware.com.
Step 8: Share and schedule
A dashboard nobody opens is just a bookmark. In Looker Studio, use Share, then Schedule delivery, to email a PDF weekly or monthly. I would send it monthly to leadership and weekly to the working team.
For agencies, a page per client with the logo and a short "what changed" text box works better than any extra chart. Write two sentences each month yourself. That is what clients read.
Common mistakes
- Tracking too few prompts. A handful of prompts gives you anecdotes. Promptwatch's own data notes you should measure across many prompts because a single one says little.
- Mixing prompt types in one average.
- Reporting only whether you were cited. The data also says to track position in the response and how ChatGPT describes you.
- Changing the prompt set every month. Version it. If you must change it, add a version column and annotate the chart.
- Ignoring model changes. When ChatGPT updates, your numbers can move for reasons unrelated to your work. Annotate major release dates on the trend line.
- Treating refresh as live. Sheets and BigQuery sources have cache and refresh settings in Looker Studio. Check the data freshness option before you tell anyone numbers are current.
Which data route should you pick?
| Route | Cost | Effort | Reliability | Best for |
|---|---|---|---|---|
| Manual Google Sheets log | Free | High, weekly | Low, small samples | Pilots and learning |
| Tracker export or connector | Subscription | Low | High, daily runs | Most teams and agencies |
| BigQuery pipeline | Warehouse + tooling | Medium to high | High | Joining with CRM or revenue |
If you are unsure, start with a tracker export and a single-page dashboard. Move to BigQuery only when you hit a real limit.
A starter checklist
- Write the four questions the dashboard must answer.
- Build a fixed prompt set of 50 or more, tagged organic, brand and comparison.
- Pick a data feed and run it for two weeks before building charts.
- Create the data source, fix field types, add the two calculated fields.
- Build scorecards, trend, segmentation and gap list.
- Blend in GA4 AI referrals and Search Console.
- Schedule a monthly email and write two sentences of commentary.
The dashboard itself takes an afternoon. The prompt set and the discipline to keep it stable are what make the numbers worth looking at.
To see how AI engines are behaving across the market while you set up your own benchmarks, the research hub at Promptwatch Data is updated regularly and is useful for sanity-checking your trends.