Key takeaways
- ChatGPT does not have a stable ranking. A SparkToro study found less than a 1-in-100 chance that two responses give the same list of brands, so single-run position is noise.
- The metrics worth tracking are aggregated over many prompts and runs: mention rate, citation rate, share of voice, co-citation, sentiment, source types, AI traffic, and a few others covered below.
- Four common numbers (a single "rank", one-off screenshots, a lone visibility score, and branded-prompt mention rate) mostly flatter you without telling you what to do.
- Read metrics in combinations. Each combination points at a different problem: coverage, prominence, crawlability, or reputation.
Why "rank" is the wrong word
Google gives you a position. ChatGPT gives you a paragraph, sometimes a list, and a handful of citations that change from run to run. The January 2026 SparkToro study had 600 volunteers run 12 prompts through ChatGPT, Claude and Google AI, each 60 to 100 times (2,961 runs total). The chance of getting the same list of brands twice was under 1 in 100. The chance of the same order was about 1 in 1,000. List length ranged from 2 or 3 items to 10 or more.
One caveat. The study was run with Gumshoe.ai, an AI tracking vendor, though SparkToro's author said he started out skeptical. Take it as a strong signal, not gospel. It still matches what anyone sees when they re-run a prompt five times.
The practical conclusion: stop asking "what is my rank?" and start asking "how often, how prominently, and how favourably do I show up across many prompts?" Every metric below is an answer to a version of that question.
The 12 metrics worth watching
1. Mention rate
The percentage of tracked prompts where ChatGPT names your brand. This is the base metric, and nearly every tracker reports it. It is only useful when the prompt set is big enough and includes non-branded prompts (more on that in the vanity section).
2. Citation rate (self-citation)
How often ChatGPT links to your own domain as a source. Mentions without citations usually mean the model knows about you from elsewhere but isn't pulling from your pages. That points at content or crawlability, not awareness.
3. Share of voice
Your mentions divided by all brand mentions across the prompt set, competitors included. It is more honest than raw mention rate because it moves when competitors move. If your mention rate is flat and your share of voice drops, someone else is gaining.
4. Position within the answer
Where you appear in the response: first recommendation, middle of a list, or an afterthought. Given the SparkToro ordering results, do not trust position from one run. Averaged over dozens of runs per prompt, it becomes a prominence signal. Promptwatch's unique domains per response data makes the same recommendation: track where you appear and how ChatGPT describes you, not only whether you're cited.
5. Co-citation and co-mention
Which competitors show up in the same responses as you. This tells you who ChatGPT considers your peer set, which is often different from who your sales team calls a competitor. It also shows which comparison prompts you should be creating content for.
6. Sentiment
Whether ChatGPT's description is positive, neutral or negative, and what it actually says (cheap, enterprise-only, hard to set up). SparkToro did not measure this, which is exactly why it is worth adding on your side. A brand can have a high mention rate and still be recommended with a caveat that kills the deal.
7. Citation source types
What kinds of pages get cited: product pages, landing pages, listicles, docs, news, Reddit, reviews. Promptwatch Data (Sep 28 to Oct 4, 2026) shows product pages as the most cited content type at 22.4%, followed by landing pages at 16.8% and listicles at 13.6%. Product and landing pages together make up 39% of citations. If your cited pages are all blog posts while the market is citing product pages, you have a format gap.
8. Citation durability
How long a cited page stays cited. Promptwatch tracked every page ChatGPT cited on July 6, 2026 for 90 days and found that landing pages and docs lasted longest: 78% of landing pages were still cited after 30 days. News lost half its pages within 36 days, and almost a quarter were never cited again after day 0. If your visibility depends on news-style content, expect to keep refreshing it.
9. Prompt coverage
The share of your target prompts and topics where you appear at all, split by prompt type: organic, competitor comparison, brand-specific. A gap here is a content brief waiting to be written. Tracking the unmentioned prompts is more useful than polishing the ones you already win.
10. Query fan-out match
ChatGPT does not search for your user's literal prompt. It generates sub-queries. Promptwatch's query fanout data shows average searches per response and query length shifting over time, and on August 8, 2026 ChatGPT began using the site: operator at scale, jumping from about 0.4% to about 17% of fanout queries. Check the fanout report's own figures before quoting its averages, since the page mixes dates. The point for tracking: your content has to match the generated sub-query, so log the fan-outs, not just the prompt.
11. AI crawler activity
Whether GPTBot, OAI-SearchBot and ChatGPT-User are actually reading your pages, and whether they hit errors. A rank tracker cannot see this because it only watches outputs. Server-side logs (or a platform that ingests them) tell you if a visibility problem is really an access problem. Promptwatch Data on AI crawler traffic gives weekly verified request benchmarks to compare against.
12. AI referral traffic and conversions
Visits and conversions from chatgpt.com and other AI sources, tied to landing pages. This is the metric your boss will eventually ask for. It's small compared to Google for most sites and often converts better, so track it by page rather than as one lump.
The 4 vanity metrics to ignore
A single "rank" number
If a tool says you rank #3 in ChatGPT for a prompt, ask how many runs that came from. With under 1-in-100 odds of repeating a list, a one-run position is a coin toss with a label. Seer Interactive makes a similar argument: there is no stable ranking, so you can't assume a click-through rate attached to a position.
One-off screenshots and spot checks
Checking one prompt once is the most common mistake. A single answer can't represent a category or a trend. Screenshots are fine for a slide, not for decisions.
A lone composite visibility score
A 0 to 100 score can be useful as a trend line, but on its own it hides which problem you have. A score of 62 could mean great mentions with terrible sentiment or the reverse. Always break it down into the component metrics above before reporting it.
Mention rate on branded prompts
Prompts like "what is [your brand]?" will always mention you. Promptwatch Data shows brand-specific prompts cite far fewer domains (only 21.3% cite 10 or more, versus 52.5% for organic prompts). Tracking mostly branded prompts overstates how often new buyers meet you. Weight your prompt set toward organic and competitor-comparison prompts.
Reading metrics together
One metric rarely diagnoses anything. Pairs do.
| Pattern you see | Likely problem | What to check |
|---|---|---|
| Low share of voice, low mention rate | Coverage | Prompt coverage, content gaps, comparison pages |
| Mentioned, but late or averaged low position | Prominence | Citation source types, competitor co-citation |
| Mentioned, few self-citations | Content or crawlability | Crawler logs, page formats, citation durability |
| High mentions, weak sentiment | Reputation | Sentiment text, review sites, third-party sources |
| Good citations, little traffic | Attribution or intent | AI referral traffic by page, prompt types |
| Sudden overnight drop across everything | Model or retrieval change | Industry-wide data, not your site |
That last row matters. Promptwatch Data showed average citations per response dropping around the GPT-5.3 rollout on March 4, 2026, and reddit.com's share of ChatGPT citations fell from roughly 4% to 0.5% on August 14, 2026. Neither was caused by anything a brand did. Before you panic or rewrite your site, check whether the whole market moved.
How to set up tracking that holds up
- Build a prompt set of 50 to 150 prompts across organic, comparison and branded types. Keep branded under a third.
- Run each prompt repeatedly, daily if the tool allows, and report averages and ranges, not single answers.
- Segment by country and language. The same prompt differs by locale.
- Log citations with exact URLs so you can see which pages and content types win.
- Pair output metrics with input metrics: crawler logs and AI referral traffic.
- Review weekly, report monthly. Daily wobble is normal.
One more caution on tool choice. Manual testing falls apart past about 10 prompts, and API-based tracking can differ from what users see in the real interface. Ask vendors whether they query the actual ChatGPT UI, how many runs back each number, and whether they log citation URLs.
Tools that can track these metrics
The category is crowded, so match the tool to the metrics you care about rather than the longest feature list.
Promptwatch covers most of the 12 in one place: real-UI prompt tracking with volumes and difficulty, Share of Voice, position and sentiment, query fan-outs, IP-verified AI crawler logs, citation analytics, and visitor analytics for AI-referred traffic. It also goes past monitoring with content gap analysis, the Content Agent and an Action Board, which matters because metrics 9 and 7 only pay off if you act on them.
Other options worth a look depending on budget and focus:
Profound is an enterprise-leaning option with deep monitoring.
Peec AI is a lighter prompt tracker with suggestions.
Otterly.AI is an affordable entry point for brand visibility monitoring.
Ahrefs Brand Radar suits teams already living in Ahrefs, though prompt sets are less flexible.
Semrush's AI Visibility Toolkit is the equivalent if your team is on Semrush.
| Tool | Strongest on | Watch out for |
|---|---|---|
| Promptwatch | End-to-end: tracking, crawler logs, traffic, content actions | More depth than a solo marketer may need on day one |
| Profound | Enterprise monitoring | Higher price point |
| Peec AI | Simple prompt tracking | No crawler logs or content generation |
| Otterly.AI | Low-cost monitoring | Basic feature set |
| Ahrefs Brand Radar | Fits an existing SEO stack | Shallower AI-specific data |
| Semrush AI toolkit | Fits an existing SEO stack | Fixed prompt sets |
For a wider shortlist, the AI rank tracking directory lists many more, and buyer guides like Omnia's 19-tool roundup include procurement checklists worth borrowing. Confirm pricing on each vendor's site, since it changes often.
A simple weekly routine
Monday: check share of voice and mention rate against last week, then confirm any big swing isn't a market-wide shift. Midweek: look at the prompts where a competitor appears and you don't, and turn two of them into content briefs. Friday: review crawler errors and AI referral traffic by landing page, and refresh any page whose citations are decaying.
That's about an hour once the prompt set exists. It beats a dashboard full of single-run ranks every time.
Bottom line
Track aggregated, repeatable metrics: mention rate, citation rate, share of voice, position, co-citation, sentiment, source types, durability, prompt coverage, fan-out match, crawler activity, and AI traffic. Ignore single-run ranks, screenshots, lone composite scores, and branded-prompt mention rates. If a number can't change what you do next week, it belongs on the vanity list.