ChatGPT rank tracker

A ChatGPT tracker records whether ChatGPT names your brand, how early in the answer you appear, which competitors are named beside you, and which domains the answer cited. Everything below explains how to measure that — first by hand, for nothing, and then with a tracker when the manual version stops scaling.

§What does a ChatGPT tracker measure?

Five fields, recorded per prompt, per run. A tool that reports one blended “visibility score” has thrown away the four fields you would have acted on.

• Inclusion. Was your brand named in the answer at all? This is the floor. Everything else is a refinement of a mention that already exists.
• Position in the answer. ChatGPT replies in prose, so there is no numbered results page to read. The workable proxy is where your first mention falls within the response — opening recommendation, mid-list option, or a hedge near the end. SimplyRank converts that offset into ten ordered bands so it can be charted.
• Cited domains. Which sources did the answer attach? Links and numbered reference lists are pulled out as URL, domain, title and the order they appeared, then flagged as your domain, a competitor’s, or a third party’s.
• Sentiment. “X is the standard choice” and “X exists, but it is expensive” are both mentions. A tracker that cannot separate them will report a good week when you had a bad one.
• Competitor set. Who else was named, and in what order? This is the field that turns brand monitoring into a market-share reading.

Note what is missing from that list: traffic, clicks and impressions. ChatGPT does not publish them, and no tracker can infer them from an answer. Anyone showing you AI “traffic” is modelling, not measuring.

§How to track your rankings on ChatGPT over time

Tracking over time is a discipline problem before it is a software problem. Three things have to stay still so that the fourth can move:

1. A fixed prompt library. Write the prompts once, in the words your buyers would use, and stop editing them. A prompt list that drifts makes every comparison meaningless, because you can no longer tell a change in ChatGPT from a change in your own question.
2. A fixed brand and competitor set. Register aliases too — the product name, the legal entity, the old name people still use. Otherwise a rename reads as a collapse in visibility.
3. A fixed cadence. Same day, same rhythm. Weekly is enough for most categories, for reasons covered below.

Then read the trend, never the single run. A ChatGPT answer is generated, not retrieved, so the same prompt can come back with a different order, a different brand list and different wording minutes later. A run of scans on stable prompts supports a claim about movement; one screenshot supports nothing.

§Run the manual ChatGPT check yourself

You do not need a subscription to start. Here is the weekly check we ran by hand before we built the scanner, and it still works as a way to decide whether the problem is worth tooling.

Ask five prompts, in a fresh chat each time. Replace the bracketed parts with your own category, buyer and competitor:

• “What are the best [category] tools for [buyer type]?”
• “I am a [role] at a [company type]. Which [category] tool should I choose, and why?”
• “What are the best alternatives to [competitor]?”
• “[Competitor] vs [your brand] — which is better for [use case]?”
• “Is [your brand] a good choice for [use case]? What are its weaknesses?”

Record six columns in a spreadsheet, one row per prompt per run: the date; the exact prompt text; whether your brand appeared; roughly where it appeared (first sentence, middle of the list, or a closing caveat); every other brand named, in order; and any links or sources the answer offered. Add a seventh column for the tone of the sentence that mentions you, in your own words.

Two rules that make the results comparable. Start a new chat for every prompt, because the previous answer conditions the next one. And use a temporary chat, or an account with memory and custom instructions switched off — otherwise you are measuring your own profile, not your category.

Repeat each prompt three times in the same sitting. If your brand appears twice out of three, that is a far more honest reading than a single lucky run, and it will teach you quickly how noisy this surface is.

Why manual checks mislead

The manual method is genuinely useful and genuinely limited. Four things go wrong with it:

• The model is not deterministic. Identical prompts produce different answers. A brand that disappears from one run has not necessarily lost anything.
• Your account is not neutral. Memory, custom instructions, past conversations and account region all steer the answer. Your ChatGPT already knows who you work for.
• The surface varies. Whether the answer browsed the live web changes the sources it can cite, so two runs can be built from entirely different evidence.
• Sample size and honesty. A handful of prompts, run by the person who most wants a good result, is the exact setup in which people stop recording the runs where they lost.

Do the manual check for a month. If the answers matter to your pipeline, the spreadsheet will make the case for automating it better than any vendor page can.

§What SimplyRank’s ChatGPT scan actually does

The automated version is the manual method with the sampling problems engineered out. Each scheduled run does the following:

• Fans every prompt out to four assistants. The same prompt text goes to GPT-4o, Claude, Perplexity and Gemini in one pass, under one neutral system prompt — the fan-out behind the cross-model AI rank tracker. The prompt is never rewritten to mention your brand, because that would manufacture the mention we are trying to measure.
• Matches brands and aliases in the response. Your brand and every competitor you registered are matched against the answer text, so a competitor list is produced from the same response rather than from a separate query.
• Derives position from the first mention. The offset of your first mention within the answer is mapped to one of ten ordered bands, so “named in the opening recommendation” and “named in a closing caveat” are different numbers on the chart.
• Extracts citations into domains. Markdown links and numbered reference lists are parsed into URL, domain, title and order, then labelled as your domain, a competitor domain, or neither. Over weeks this becomes a ranked list of the pages your category’s answers are built from.
• Classifies sentiment per brand. A separate model reads the response and scores each brand it names from negative to positive, with one sentence of reasoning attached so the score can be argued with.
• Stamps a benchmark version on the session. Each session stores the models used, the target market and a hash of the exact system prompt that was sent. When we change the instrument, the change is labelled in the data instead of appearing as a mysterious swing in your visibility.
• Keeps one canonical session per brand per day. Re-running a scan does not inflate a trend line, and a run in which providers failed on a meaningful share of prompts is excluded from analytics rather than averaged in.

On the model itself, we are deliberately plain: ChatGPT scans call GPT-4o through the OpenAI API. OpenAI’s own model documentation, checked on 28 August 2026, lists dated snapshots behind that model name and marks one of them as the current default, and notes that pinning a snapshot is what keeps behaviour consistent. Naming a model is therefore not a stability guarantee on its own — the benchmark version stamped on each session is what lets you see when the instrument changed underneath a trend.

Why weekly beats daily

Daily scanning of a generative model mostly measures the model’s own variance. You watch ChatGPT express the same recommendation three different ways and read it as movement. A weekly gap is long enough for a real cause — a new comparison page, a competitor launch, a model update — to separate from that noise, and short enough to catch it in the month it happens. Daily cadence is available from the Starter plan up, and it earns its place during a launch or an incident; as a standing habit it costs attention and buys very little.

What ChatGPT rewards, and what it ignores

§What a ChatGPT tracker cannot do

Worth stating before you buy anything, from us or anyone else:

• It cannot see individual users’ chats. Scans run in clean, stateless API sessions. That is a feature — it removes personalisation — but it means the measurement is of a neutral session in your target market, not of the conversation your actual buyer had.
• It cannot guarantee a ranking. No vendor controls what a model recommends. A tracker tells you where you stand and what the answer cited; the work of changing it is content, positioning and third-party evidence, and it can fail.
• It cannot make a single answer meaningful. Because responses vary between runs, we report trends across a fixed prompt library. Any single scan — including a flattering one — is a sample.
• It cannot cover every ChatGPT surface. Custom GPTs, and answers shaped by a user’s memory or uploaded files, are separate surfaces from the base model our scans read.
• It cannot report AI traffic. Nobody can, from the answer alone. What a scan reports is presence, position, sources and tone.

§Choosing a ChatGPT rank tracker for SaaS

For a SaaS team the deciding factor is the prompt library rather than the dashboard. Buyers do not type keywords at an assistant; they describe their situation. So the questions worth asking any vendor are mechanical:

• Can you run your own prompts, in your buyers’ language, and lock them?
• Is the raw answer stored, so you can re-read what was actually said?
• Are competitors matched inside the same response, not in a separate query?
• Can you export the rows and check the tool’s working?
• Does the same prompt library run against the other assistants your buyers use?

SimplyRank is built to that shape: your own prompts, competitor matching in the same answer, stored responses, and CSV export from the Starter plan up. Plan limits — the 14-day trial with 10 tracked prompts, and 25 to 300 prompts per scan across the paid tiers — are on the pricing page, so you can check the claim rather than take it.

If cost is the blocker, read free AI rank trackers, honestly compared before you buy anything from anyone. It lists what each free checker actually covers, with the date we checked it.

When ChatGPT names you but another assistant does not, the difference is usually in how each one weighs evidence. The Claude rank tracker and the Perplexity rank tracker cover their own citation behaviour, and Claude vs ChatGPT sets the two side by side on the same evidence.