How to Track ChatGPT Rankings Over Time

How to track ChatGPT rankings over time: what to log, how often to sample, spreadsheet vs tool, and how to read chatgpt rank tracking data.

Checking whether ChatGPT recommends you once is easy. The useful question is harder: are you gaining or losing ground? Learning how to track ChatGPT rankings over time turns a one-off curiosity into an actual feedback loop — you change something, the trend responds, you know what worked. This post covers the method: what to log, how often to sample, how to do ChatGPT rank tracking in a spreadsheet, when to switch to a tool, and how to read the data without fooling yourself.

Definition: ChatGPT rank tracking is the practice of running a fixed set of buying-intent prompts against ChatGPT on a regular schedule, logging whether and where a brand appears in each answer, and analyzing the results as trends — mention rate, average position, and share of voice — rather than as single data points. Because individual answers vary run to run, only repeated sampling produces numbers you can act on.

Why tracking beats checking

Three forces move your ChatGPT visibility without telling you:

1. Model updates. New model versions replace old ones several times a year, each with different training data and different habits. A release can reshuffle category recommendations overnight. OpenAI publishes changes in its model release notes, but never at the level of "your brand moved."
2. Index refreshes. Search-mode answers reflect the current web. A competitor's new comparison page, or a listicle update that drops you, changes answers within days.
3. Your own work. Outreach, new content, review pushes — the whole playbook in how to rank on ChatGPT — pays off over weeks. Without a trend line, you cannot attribute the payoff or catch the regressions.

A single check catches none of this. A weekly series catches all of it.

What to log on every run

Consistency of fields matters more than fancy analysis. For each prompt, each run, record:

• Date and prompt — verbatim; do not "improve" prompts mid-series or you break the trend.
• Status — Named (your brand is in the answer text), Cited only (your page is a source but the text names someone else), or Absent. Your most important field.
• Position — 1 for first brand named, 2 for second, and so on; blank if absent.
• Competitors named — the full list, in order. This funds share-of-voice analysis later.
• Description — the phrase ChatGPT used for you. Wrong or stale descriptions are findings in their own right.
• Search used — whether the answer browsed the web (visible in the interface). Memory answers and search answers trend differently, and knowing which moved is diagnostic.
• Sources cited — for search answers, which pages. These are your outreach targets, and the reason citation tracking exists as a feature.

How often to sample

In our testing, the practical schedule is:

• Per prompt: three runs per sampling day, in fresh sessions with memory off. Three is the floor at which a mention rate stops being coin-flip noise.
• Cadence: weekly for prompts tied to revenue; monthly is enough for peripheral prompts. Daily sampling is rarely worth it manually, though automated trackers do it cheaply and catch step-changes faster.
• After events: always sample within a few days of a major model release, and after you publish or land anything significant.

Twenty prompts, three runs, weekly — that is sixty answers a week to elicit and log. Feasible by hand, tedious by month two. Plan for that honestly.

The spreadsheet method

One sheet, one row per prompt-run, the columns above. Then three derived views:

1. Mention rate per prompt, per week — appearances divided by runs. Your headline metric.
2. Four-week moving average — smooths run-to-run noise so you react to real moves. A prompt swinging between 33 and 67 percent weekly is probably flat; a moving average sliding from 60 to 20 over a month is a real loss.
3. Share of voice — across all runs in a period, your mentions divided by total brand mentions, next to the same figure for each named competitor.

The spreadsheet method genuinely works and costs nothing. Its failure mode is not analytical but human: it depends on someone doing sixty manual chats every week indefinitely, and skipped weeks destroy exactly the continuity that makes trends meaningful.

The tool method

A rank tracker automates elicitation, logging, and the derived views: scheduled runs, fresh sessions, position history, share of voice, cited sources, and alerts when an answer flips. That last one changes behavior — you learn about a drop the day it happens, not the next time someone remembers to check. This is what our ChatGPT rank tracker does, with daily tracking and alerts on every plan, and the same prompt set can run across seven more engines from one place.

One honest caveat about all tracking tools, ours included: tools typically query models via API, and API answers are not always identical to the consumer app — the app layers on memory, custom instructions, and its own retrieval behavior. Trends and relative movements transfer well; treat exact positions as approximate. Any vendor claiming pixel-perfect app replication is overclaiming.

Reading the data without fooling yourself

• React to moving averages, not single weeks. One bad sampling day is weather; four bad weeks is climate.
• Segment by answer mode. If search-mode answers dropped you but memory answers still include you, the cause lives on the web (a page changed), not in the model — and it is fixable this week.
• Watch descriptions, not just positions. Sliding from "best free option" to "a basic option" while holding position two is a real loss that position tracking alone never shows.
• Treat correlation honestly. You published a comparison page and mentions rose two weeks later — that is a signal worth repeating, not proof. Multiple things change every week; the discipline is running the play again and watching the trend respond twice.
• Expect discontinuities at model releases. Annotate release dates on your charts. A step-change on a release date is a regime shift; re-baseline rather than agonizing over week-over-week deltas across it.

Check where you stand

A trend line needs a starting point. Run the free AI visibility checker — submit your domain and email, and you get a baseline report across all 8 engines, ChatGPT included: which prompts mention you, at what position, and who is winning the ones that do not. Report emailed within 24 hours (beta).