Not Every AI Engine Shows Its Sources
The three major AI engines differ so much in whether they show sources, and how many, that a single blended visibility score tells you almost nothing useful.
Key Takeaways
- Perplexity named a source on 100% of answers. Google AI Overview on 94.9%. ChatGPT on 75.5%.
- One in four ChatGPT answers has no sources at all, so there is no citation to win.
- Sources per answer: Perplexity 18.9, AI Overview 10.0, ChatGPT 4.5.
- A blended AI visibility score is dominated by whichever engine cites most generously.
- Set targets per engine, and compare against competitors on the same engine.
Why we looked into this
Companies increasingly ask a version of the same question: are we visible in AI search? It sounds like one question. It is at least three, because the engines behave so differently that a single answer is misleading.
Before comparing your performance across engines, or drawing conclusions from a report that blends them, it is worth knowing something basic that almost nobody checks. Do these engines even show sources? And do they show them at the same rate?
They do not, and the gap is large enough to change how you read every other number you have been given.
How we tested it
This one is simple, which is part of why it is useful. For every one of the 6,601 answers we recorded, we counted how many sources the engine attributed. Not whether our brand appeared. Just whether the engine named any source at all, and how many.
Because each question was asked daily for a month on all three engines, the comparison is like for like. Same questions, same days, three different behaviours.
What we found
Perplexity named a source on every single answer it gave, across the entire month. Google AI Overview did so on 94.9%. ChatGPT did so on 75.5%.
Put plainly: one in four ChatGPT answers names no source at all. For those answers there is no citation to win. No amount of content work puts you in them, because there is no list of sources for you to appear on.
The number of sources differs just as sharply. When Perplexity answers, it names about nineteen sources. Google AI Overview names about ten. ChatGPT names about four and a half.
Why this matters more than it looks
Combine those two numbers and the amount of opportunity each engine offers is wildly different. Perplexity gives you nineteen chances to appear on every answer. ChatGPT gives you four and a half, three quarters of the time.
That has an uncomfortable implication for how AI visibility is usually reported. If someone hands you a single score for AI visibility, blended across engines, it is dominated by whichever engine cites most generously. A brand can look strong in a blended report largely because Perplexity is easy to appear in, while being close to invisible on the engine its buyers actually use.
The reverse trap is just as common. A brand can look weak on ChatGPT and conclude its content is failing, when a quarter of the answers had no sources to be part of in the first place.
What this means for your business
First, insist that any AI visibility reporting is broken out by engine. A blended number is not a summary, it is a distortion. This applies to reports you receive from agencies and to tools you buy.
Second, decide which engine matters for your buyers before you set targets. If your customers research inside ChatGPT, you are competing for four and a half slots on three quarters of answers, and being quoted at all is a genuine achievement. If they use Perplexity, nineteen slots per answer means presence is more achievable and the real contest is being near the top rather than being present.
Third, when you compare yourself against a competitor, make sure you are comparing on the same engine. A competitor who looks dominant may simply be measured somewhere more generous.
None of this is a reason to ignore ChatGPT. It is the engine where citation is scarcest, which makes each citation worth more, not less. But the goal has to be set against how the engine actually behaves rather than against a number that averages three different behaviours into one.
We report per engine for exactly this reason at Novastacks.
Methodology
89 questions agreed with chief executives and business owners as ones their companies need to monitor, put to ChatGPT, Google AI Overview and Perplexity every day through August 2026, producing 6,601 answers. For each answer we recorded how many sources the engine attributed.
Perplexity coverage in this study spans a subset of the question set, so its perfect citation rate rests on a smaller sample than the other two engines. The ChatGPT and Google AI Overview figures cover the full set. All brands are anonymised.
Frequently Asked Questions
Does ChatGPT always show sources?
No. Across 2,759 answers, ChatGPT named a source on 75.5% of them. One in four answers cites nobody at all, so there is no citation available to win.
How many sources does each AI engine cite?
Perplexity averages 18.9 sources per answer, Google AI Overview 10.0 and ChatGPT 4.5.
Why is a single AI visibility score misleading?
Because the engines differ so much in citation behaviour that a blended figure is dominated by whichever engine cites most generously. A brand can look strong overall while being close to invisible on the engine its buyers use.
Which AI engine is hardest to get cited by?
ChatGPT, on this evidence. It cites least often and names the fewest sources when it does, which makes each ChatGPT citation scarcer and more valuable.
How was this measured?
89 questions were put to all three engines every day through August 2026, producing 6,601 answers. For each answer we recorded how many sources the engine attributed.
Which Engine Actually Matters for Your Buyers?
We measure your visibility separately on each engine, against the questions your buyers ask, so you can see where you genuinely stand rather than an average that hides it.