Your chances of being cited vary more than fivefold depending on which AI answers

Ask an AI a question and it answers. Sometimes it lists the sources it used; sometimes it lists nothing at all. When it does list them, it may name two or it may name fifteen.

Put end to end, over 1,949 answers measured between 6 July and 1 August 2026, the difference is large: an answer from Perplexity carries 13.1 distinct sources on average, an answer from Claude carries 2.3. A factor of 5.6 in how many chances there are for a name to appear.

That single number hides two very different behaviours, and separating them is the whole point.

AI Coverage: does the answer cite anyone at all

AI Coverage is the share of an engine's answers that name at least one source. It says whether there is a door.

An engine at 100% shows its sources every time it answers. An engine at 51.8% does it in one answer out of two, and the other half is written without crediting anyone.

AI Density: how many sources, when it does cite

AI Density is the number of distinct sources an answer carries, counted only over the answers that cite at all. It says how wide the door is.

Counting it over every answer would fold AI Coverage back into it, and the two numbers would then say the same thing twice. Counted over sourced answers only, they are independent: one measures how often, the other how many.

The two, side by side

The last column is the product of the first two, and it is the number that matters: the average count of distinct sources an answer carries.

Read the middle three rows to see why one number would not have been enough. Grok, ChatGPT and Mistral land within half a source of each other, by three different routes. Grok cites almost every time but names few. ChatGPT cites in three answers out of four and names more each time. Mistral sits between the two. A single figure would have made them look alike; they are not.

What each number is good for

AI Coverage answers a question about opportunity: on how many of the questions your market asks does an engine open its answer to outside sources at all. It moves slowly and it is a property of the engine, not of the subject.

AI Density answers a question about room: when the door opens, how many names go through it. An engine at 13 leaves space for a long list; an engine at 4 or 5 keeps a short one, and the competition inside it is correspondingly harder.

Neither replaces the other, and their product is the only figure that should be compared across engines without further thought.

Method

The measurement covers the live measurement panels running on the Epovest platform between 6 July and 1 August 2026, a total of 1,949 answers across six engines. Two of the six entered the panel on 24 July and are therefore measured over a shorter window.

A source is the registrable domain of a page the engine cites, resolved against a versioned public suffix list, so that three pages of the same site count as one source. Only pages the engine cites in its answer are counted; pages it reads without crediting are excluded, since only one engine exposes them.

Every engine was queried through its API on the fast tier it serves by default, and each reported a single model over the whole window, so no model rotation sits inside these figures.

The questions come from live panels dominated by leisure travel, local professional directories and business software discovery, in six languages.

One more thing

The sources behind those citations are not the same ones from one engine to the next. Compared question by question, on the same day, two engines share between 2.1% and 19.1% of the sources they name.

So the factor of 5.6 is not a gap between more and less of the same thing. The engines that cite the most and the engines that cite the least are not drawing on the same web at all. That result deserves its own study, and it will get one.