AI NativeNot every AI answer is built the same way. Some answers are grounded: the engine searched the web before responding and pulled specific documents into its reasoning. Other answers are parametric: the model answered from the knowledge baked into it during training, without fetching anything at the moment of the question. That difference has direct consequences for how you appear, and for what you can do to change it.
Grounded answers
A grounded answer is one where the engine performed a web search first. The answer then reflects what is currently on the web, not just what a model absorbed during training. Grounded answers come with citations: the sources the engine read before responding.
Those citations are auditable evidence. They tell you which domains the engine trusted enough to draw on, which means they tell you where the authority that shaped your recommendation came from. When you are recommended in a grounded answer, there is a chain of reasoning from a source you can identify to the sentence that put you forward. When you are absent from a grounded answer, the sources the engine chose did not carry enough of the right evidence about you.
The citation panel in the scan results shows you which domains appear in answers where you rank at each ladder level. That is the map of which sources you need to influence if you want grounded answers to represent you better.
Parametric answers
A parametric answer draws on what the model knows from training. There is no web search, no citation list, no verifiable chain. What you get is the model's internal representation of your brand and your category.
For large, well-documented brands, parametric knowledge can be reasonably accurate. For newer brands, niche products, or anything where the training data was thin or outdated, parametric answers are more likely to be incomplete, dated, or absent. There is also no clear path to changing them: parametric knowledge updates slowly, and only when a model is retrained on new data.
This is why grounded surfaces are the primary focus of the scan. On a grounded surface you can trace what is driving the answer, and you can change it by changing the sources. On a parametric surface you are mostly waiting.
How AI Native reads provenance
When a scan runs a grounded prompt and the engine returns citations, each citation is classified as grounded. If an answer returns without citations on a surface that is capable of searching, it is more likely to be parametric. The provenance_class field on each citation records which kind it is.
The provenance classification is not a judgment about whether the answer is good or bad. A parametric answer can still recommend you. A grounded answer can still get you wrong. What provenance tells you is which kind of intervention is available: source-side work for grounded answers, entity and schema work for parametric ones.
What citations tell you about accuracy
Grounded answers can be checked against reality. When an answer cites your site and then makes a specific numeric claim about your product, AI Native can compare that claim against the verified facts you have entered in the Brand Truth Studio. A mismatch is flagged as a potential accuracy problem, with the claimed number, the truth, and the source URL recorded in the detail row.
Parametric answers that make numeric claims can also be checked, but when a number is wrong in a parametric answer there is no citation to point to. The answer is wrong because the model's training data was wrong or out of date, and fixing it is a slower, less direct process.
See Accuracy and fact checks for how the accuracy audit works in detail.
Questions
What does it mean when an answer has no citations?
It usually means the engine answered from model memory rather than from a live web search. The answer is parametric: it reflects what the model absorbed during training, without fetching current sources. Some engines always search; others have a search mode and a non-search mode. The scan notes which mode was active for each answer.
Can I appear in grounded answers if my site is not cited?
Yes, if your brand name or a recognised alias appears in the answer text. The presence check is text-first; citations are additional evidence. Being cited by your own domain is the strongest form of citation-based presence, but other paths exist. Category sources, review sites, and third-party comparison pages can all carry your brand's name into an answer.
Do all AI engines return citations?
No. Some engines return grounded answers with full citation lists. Others operate parametrically and return no citations. A handful support both modes. The scan runs on the surfaces configured for your product and records what each one returned. You can see the per-engine citation coverage in the answer detail.
Why does the same domain sometimes appear in answers where I am absent?
A domain being cited does not mean your brand name appeared in the answer. A third-party category guide might be cited and might mention several competitors without mentioning you. Citation presence indicates what the engine read; brand presence indicates whether what it read included you. They are related but not the same thing.
How do grounded citations connect to what I should work on?
The citation panel shows you which domains appear most often in answers where you reach each ladder rung. The domains in answers where you are recommended are the sources already working in your favour. The domains that appear in answers where competitors are recommended but you are absent or merely mentioned are the sources that do not yet carry enough of your brand's story. That is the priority list for distribution and coverage work.
Where can I learn more about the accuracy audit?
See Accuracy and fact checks. For the ladder positions these citations feed into, see The visibility ladder.