How to track ChatGPT rankings over time

Why there is no ranking position in ChatGPT

A Google results page is an ordered list. Ten blue links, numbered, in sequence, and everyone can agree what position 3 means. An assistant answer is not a list. It is generated text that may name a few products, cite a handful of sources, or cite nothing at all, and the composition changes between two people asking the same thing an hour apart.

Three things make a stable position impossible. The answer is generated fresh each time, so there is randomness baked in. It is personalized by the conversation before it, meaning the same question asked after a different opening gets a different answer. And the phrasing of the prompt changes the result far more than a keyword variation changes a Google page, because you are talking to it rather than querying it.

So when somebody asks how to track their ChatGPT rankings over time, the useful reframing is: pick the prompts your buyers would actually type, run them repeatedly, and count how often you show up. The metric is a rate of appearance across a sample, not a position. Treat it like a poll rather than a scoreboard.

What the AI visibility tools actually measure

A category of tools now exists for exactly this job, including Rankscale, ZeroRank and Rankshift AI, and both Semrush and SE Ranking have added AI visibility modules to the dashboards their customers already pay for. They differ in coverage and price, but the mechanism underneath is the same one in all of them.

They keep a list of prompts you define, send those prompts to the assistants on a schedule, and parse each answer for your brand. From that they derive the numbers you see: how often you were mentioned, how often you were cited with a link, which competitors appeared alongside you, and a share of voice figure comparing all of it. Run the same prompt set weekly and you get a trend line.

Two honest caveats before you buy one. Sampling error is real, because a handful of runs per prompt against a generative system produces noisy numbers, and a change of a few percentage points week to week may be nothing at all. And the tools cannot see the traffic, only the mention. Assistants pass through very little referrer data, so knowing you were cited is not the same as knowing anyone clicked.

How to track AI Overview rankings over time

This one is tractable, and it is where most of the money still is. An AI Overview appears on a normal Google results page above the organic results, and it is drawn from pages that already rank. That means the surface behaves enough like classic search to be measured properly: the same query, in the same location, on the same device, checked daily.

What you want recorded on each check is whether an overview was present for the query at all, whether your domain was cited inside it, and what position you held in the organic results underneath it. That third one matters more than people expect. BrightEdge put AI Overviews on 48 percent of the queries it monitors as of February 2026, up 58 percent year over year, and Ahrefs has measured the top organic result losing roughly 58 percent of its clicks when an overview sits above it. A position you held all year can be worth substantially less than it was without moving at all.

Which is the argument for tracking presence as a time series rather than checking it once. A keyword that gained an overview in March is a different keyword commercially than it was in February, and the only way to know the date that happened is to have been recording it. Our page on AI Overview tracking covers how we record presence alongside the daily position beneath it.

What does Search Console show about AI Overviews and AI Mode?

Google launched a generative AI performance report in Search Console on June 3, 2026, which broke AI surface data out from standard organic data for the first time. It reports impressions, pages, countries and devices. It does not report clicks, click through rate or query data. It is worth opening, and it is worth knowing its limits before you build a process on it.

The significant limitation is that it groups every generative surface into a single view. AI Overviews, AI Mode and the generative features in Discover are all counted together, and there is no way to isolate which of the three produced a given impression. Google also confirmed this is a breakout rather than new data: those impressions were already inside your overall totals, so your aggregate numbers did not change when the report appeared.

Practically, the report tells you that AI surfaces are showing your pages and roughly how much. It cannot tell you which query triggered an overview, whether you were cited in it, or what it did to your clicks. For that you need per keyword checks, which is the same gap that exists between Search Console average position and a rank tracker on the classic side.

Does ranking first on Google get you cited by ChatGPT?

It helps and it does not guarantee anything. Assistants that browse the live web lean on search results to find sources, so pages that rank well are more likely to be retrieved and quoted. But retrieval is only the first step. What gets quoted from a retrieved page is the part that reads as a clean, self contained answer, which is why a well structured page at position 6 routinely gets cited over a rambling one at position 1.

That is the actionable part of all this. Measurement is worth setting up, but the work that changes whether an assistant quotes you happens on the page: answering the question in the first two sentences, putting comparisons in real tables, dating your content, and rewriting the thin sections of pages you already published so each one states a fact rather than circling it. None of that is new advice. It is just that the payoff now shows up in two places instead of one.

How often should you check AI visibility?

Monthly is enough for assistant mentions, and weekly is the most that is defensible. The measurement is noisy enough that daily readings would show you sampling variance and you would misread it as movement. Pick a fixed prompt set, leave it alone, and compare like with like over a quarter.

AI Overview presence is the opposite case and belongs on a daily cadence, because it attaches to a specific query and changes on a specific date. If an overview appears on your best commercial term, you want that dated, for the same reason you want a rank drop alert dated. A month later, nobody can reconstruct when it started.

Do you need a separate tool for ChatGPT rank tracking?

If assistant mentions are a real channel for you, yes, and you should buy one of the dedicated AI visibility tools rather than expecting a rank tracker to do it. We will be straight about our own scope here, because the category is full of vendors implying otherwise. Serpstracker tracks Google, daily by location and device, and records whether an AI Overview was present. It does not score your share of mentions in ChatGPT, Gemini, Claude or Perplexity, and a tool that claims to do both jobs well is worth a hard look.

For most US businesses the sensible sequencing is still to get the classic measurement right first, because Google remains where the volume and the buying intent are, and the pages that earn AI citations are overwhelmingly the same pages that rank. Once daily position history and overview presence are in place, add a prompt tracker if the channel justifies the line item.

The short version

Stop looking for a position number inside ChatGPT, because there is not one. Measure mention rate across a fixed prompt set, monthly, with a tool built for it, and read the results as a poll with a margin of error. Track AI Overview presence daily against the keywords you already care about, since that is a dated event on a page you can measure. Open the Search Console generative AI report for the direction of travel, knowing it combines three surfaces and withholds clicks and queries. Then spend the rest of the time on the pages themselves, which is the only input that moves both numbers.