AI visibility tracking answers one question, repeatedly: when people ask AI engines about what you do, do the answers name you, cite your site or recommend you, and is that changing? It is for marketing leads, founders and agencies who need a figure they can defend to a client or a board, not a screenshot of one chat. The Ranking Factory asks the same questions of every engine you choose, keeps every answer, and reports each rate with the number of answers it rests on.
An AI answer is not fixed. Ask the same question twice and a different business can be named. So tracking here is built around three choices: which engines and channels, which questions, and how many times. Each is explained below.
What a run asks, of whom, how often, and how the answers are counted.
ChatGPT, Gemini, Perplexity, Claude, Grok and DeepSeek, on your own keys. You pick up to four for a run; the picker lists each model cheapest first with an estimated cost per answer. Engines are asked to search where they can: Gemini with Google Search grounding, OpenAI models with a search tool, Claude with Anthropic’s web search, Grok 4 and later with xAI’s, and Perplexity always. An answer from training is labelled as one and reported apart.
Each topic is asked one question without your name and one with it, chosen from the topic’s own words by a rule, so the same topic gets the same question every run. Every engine in the run gets every question, so their results compare directly. A run asks first about topics where work has landed and nothing has re-checked yet, then about the topics asked longest ago, so none sits unmeasured at the bottom of the list.
Three signals per engine, each with a Wilson 95% range and the count of answers behind it. Per engine you also see how many answers to questions without your name still named you, and which of your tracked competitors were recommended in answers that did not recommend you.
Each question can be asked once, three times or five times. Repeats measure stability, reported apart from the rate, with the questions whose repeats disagreed listed as counts (“named you 2 of 5”). Repeats multiply calls on your key, so the form states the cost before you choose.
Every run records which engines it used. The trend is a line on a time axis, one point per run, and it breaks where an engine was added or dropped, with the change marked and named; a new model of the same engine shows as a tick. Since 11 October 2026 every qualifying project is measured weekly by default, announced by email first with the questions, the engines and an estimate of the cost.
Up to 20 exact questions per project, asked daily, weekly or monthly, every answer kept and searchable, and kept out of the score. They can also be asked of the ChatGPT and Gemini apps themselves and of Google AI Mode, through DataForSEO on your own account, from your project’s location. Google AI Overviews are watched as their own surface. Each channel has its own line, and none is added to another.
Worked example: an illustration
The shapes are the product’s: counts out of each engine’s answers, with the Wilson 95% range the run detail shows. The business and every count are made up: a bookkeeping firm with four topics, each asked two ways, plus two questions about the firm itself, on three engines.
| Engine | Named | Cited | Recommended | Named without the name in the question |
|---|---|---|---|---|
| ChatGPT | 6 (31% to 83%) | 2 (6% to 51%) | 3 (11% to 60%) | 1 of 4 |
| Gemini | 4 (17% to 69%) | 3 (11% to 60%) | 2 (6% to 51%) | 0 of 4 |
| Perplexity | 5 (24% to 76%) | 4 (17% to 69%) | 2 (6% to 51%) | 1 of 4 |
The ranges are 95% Wilson intervals for those counts. The last column is left as counts here: four answers are too few for a percentage to mean much.
With ten answers per engine the ranges overlap almost entirely. ChatGPT’s 6 of 10 and Gemini’s 4 of 10 do not show that ChatGPT treats the firm better; a difference means something only when the ranges stop overlapping. To narrow them, add questions, not engines: more engines give a fuller picture, not a sharper one.
The last column is the one that says whether an engine reaches for the firm unprompted. The named column includes questions that carried its name, so it is never presented as being found on its own.
Every number on screen states what was measured, how much of it, and on which channel. These are the limits that follow.
The rates are what the engines answer when asked. No AI engine reports how often a person saw an answer naming you. The only impression counts in the product are Google’s own, from the Search Console export you upload, and they are kept apart from everything measured.
The score comes from each engine’s API, which can be repeated and given a range. The ChatGPT app a person types into can answer differently, which is why the apps and Google AI Mode are asked separately, only for tracked prompts you switch on, and never added to the score.
Ten questions per engine gives ranges about fifty points wide. That is shown rather than hidden: before a run, the form says how firm a rate from that many questions will be. With 25 questions, a 20% rate on one engine could really be anywhere from 9% to 39%.
Two runs on the same engines can ask about different topics, and the score moves when the mix of questions with and without your name moves. Screens that compare runs say so, and the trend marks days when questions were reworded or replaced.
A rate that rises after you publish is reported as observed. No control group exists, and answers vary from run to run.
[SCREENSHOT: /tools/visibility-audit, showing an opened tracked prompt: its answer strip, one cell per answer by date with a row per engine and app (cited, named, not named, no answer)]
AI visibility tracking is on every plan, on your own AI keys, with up to 4 of the 6 engines per run and each project measured weekly by default once it qualifies. Growth monthly starts with a 14-day free trial; annual billing is ten months for twelve.
$49 a month
Visibility runs, tracked prompts and weekly measurement. AI Overview watch with a $10 monthly default budget.
$129 a month
Everything in Starter, automatic topic discovery, and a $40 monthly AI Overview default budget. 14-day free trial.
$249 a month
Everything in Growth for many clients, with white-label reports and a $40 AI Overview default budget.
$499 a month
Everything in Agency, with an AI Overview budget you set.
Engine calls are on your own keys. On our own account in September 2026 a weekly run of 10 questions on two or three engines with live search cost $0.55 on average, about $2.40 a month per project. Repeats multiply that, and the form says by how much.
Google AI Overviews, the ChatGPT and Gemini apps and Google AI Mode run on your own DataForSEO credit. The apps and AI Mode cost about $0.0012 a question, about 10 cents a month each for 20 weekly prompts, outside the AI Overview budget.
The same product, from another task
Every rate with its range and the answers behind it, every channel on its own line, and a trend that never joins two different methods.
Start your 14-day free trial