Key Takeaways
- Treat a cited answer and an uncited answer as two different machines. One was written from documents opened seconds ago. The other was written from what the model absorbed months ago. They fail in opposite directions and you fix them in different ways.
- Assume the uncited answer is where your reputation actually lives. It is faster, it runs far more often, and there is no link for anyone to check. If it describes you wrongly there is nothing to correct except what the model reads next time.
- Get quotable if you want to appear in the cited half. Grounded answers are assembled from pages that were open. A page that states a plain checkable fact in one sentence gets lifted. A page that buries it in a brochure paragraph does not.
- Get corroborated if you want to appear in the uncited half. Nothing gets absorbed from a page nobody else repeats. Directories, roundups, interviews and write-ups on domains you do not own are what put a description of you in general circulation.
- Check both before you spend anything. Ask the same buyer question in a tool that shows sources and one that usually does not. Where the two disagree tells you which of the two problems above you actually have.
Ask an AI assistant a question and you sometimes get an answer with small numbered links beside it. Ask a different question and you get the same confident paragraph with nothing attached at all.
That difference has a name, and it is the single most useful piece of plumbing a business owner can understand about how these tools work. The word is grounding.
It explains why you can sometimes find the page that described you and sometimes cannot find anything at all. It also explains why the fix is not one thing but two.
What is grounding in AI answers?
Grounding is when an assistant fetches real documents at the moment you ask and writes its answer from those documents instead of from memory. Because it opened specific pages, it can list them, and you can click through and check the claim yourself.
An ungrounded answer is written from what the model absorbed during training. Nothing was opened. There is no page behind the sentence, so there is nothing to cite and nothing to click.
Both kinds of answer look identical to your buyer. Same tone, same confidence, same three names. The only visible difference is whether links appear beside them, and most buyers do not notice either way.
Why some answers cite sources and others do not
The tool decides, per question, whether it needs to go and look. Questions about current facts, prices, availability, or anything the model suspects has changed tend to trigger a fetch. Questions that feel like general knowledge usually do not.
Google describes this openly in its documentation on AI features in Search, where generated responses are assembled from what is already indexed and the supporting pages are shown alongside. The mechanism there is retrieval first, writing second, which is the same shape covered in what retrieval means in AI answers.
Perplexity sits at the other end of the range. Its help center describes a product built around searching and citing by default, which is why it shows sources on almost everything. ChatGPT sits in between: OpenAI's explanation of what ChatGPT is describes a model that answers from training and browses when the question calls for it.
So the honest answer to why one buyer saw links and another did not is that they asked slightly different questions of slightly different products on different days. That variation is normal, and it is covered in more detail in why different AI tools give different answers.
Answered from what the model already absorbed
Answered from documents fetched while you wait
The lower route is slower because it stops to open pages. That stop is the whole difference: it is what produces a link, and it is the only route where a page you publish can be read on the day it is needed.
What grounding changes for your business
It splits your visibility problem into two problems that have nothing in common except the symptom. In the grounded half you are competing to be quoted from a page. In the ungrounded half you are competing to be part of what the model absorbed in the first place.
The grounded half moves fast. Publish a clear page today and a tool that fetches can find it this week. The ungrounded half moves slowly and on somebody else's schedule, because it only changes when the model is next trained on a web that describes you differently.
That asymmetry is why owners get confusing results. A coach publishes a good page, sees themselves cited in Perplexity within days, and concludes the problem is solved. Then a client mentions that ChatGPT gave them three other names, because that answer never opened anything.
| Grounded answer | Ungrounded answer | |
|---|---|---|
| Where the words come from | Pages opened while the buyer waits | What the model absorbed in training |
| What the buyer sees | An answer with links beside it | The same answer with nothing beside it |
| Can you check the claim | Yes, by opening the cited page | No, there is nothing to open |
| How fast a new page reaches it | Days, once the page is indexed | Months, and only at the next training pass |
| What moves the result | One quotable sentence on a page it can open | Many independent sites describing you the same way |
| If it gets you wrong | Fix the page it cited | Publish corroboration and wait |
The two problems, and which one you have
You can tell them apart in ten minutes. Ask the same buyer question twice: once in a tool that shows sources on nearly everything, once in a tool that usually answers directly.
Named in the cited answer but not the direct one means your pages are fine and your corroboration is thin. Something readable exists, but not enough of the web repeats it for you to have been absorbed.
Named in the direct answer but not the cited one is rarer and usually means your site is hard to lift from. You are known, but no single sentence on your pages is clean enough to quote.
Named in neither is the common case, and it is the honest starting point for most small businesses. It is also the least alarming of the three, because it usually means nobody has published a plain description of what you do anywhere, including you.
What to do about each of them
For the grounded half, write one page per real buyer question and answer it in the first two sentences. Google's guidance on creating helpful content is the closest thing to an official statement of what makes a page worth lifting from, and it has not changed much: say the useful thing plainly, early, and in your own words.
Put the checkable facts in single sentences. What you charge, who you work with, where you work, how long it takes. A grounded answer is assembled from fragments, so a fragment that answers a question on its own is worth more than a beautiful paragraph that only works in context.
For the ungrounded half, get described somewhere you do not own. An interview, a directory entry, a roundup, an association listing. Repetition across independent domains is what puts a description into general circulation, which is the same argument made in what it means when AI cites a website.
One honest limit. Nobody outside these companies can tell you exactly when a tool decides to fetch, and the rule changes without announcement. What you can control is whether there is something worth quoting when it does, and whether anyone besides you has said it.
If you want that measured rather than guessed at, the free check asks ChatGPT 25 real buyer questions from your niche at $0 and emails a plain-English readout. It runs as soon as you ask for it, and the report lands within the hour between 7am and 5pm Central, or by 10am the next morning outside those hours.
Questions we hear the most
What is grounding in AI answers?
Grounding is when an assistant fetches real documents at the moment you ask and writes its answer from those documents instead of from memory. Because it opened specific pages, it can list them, and you can click through and check the claim.
Why do some AI answers cite sources and others do not?
Because only some answers were grounded. When the tool decides a question needs current or verifiable information it goes and looks, then cites what it opened. When it decides the question is general knowledge it answers directly, and there is nothing to cite.
Is a grounded answer more accurate than an ungrounded one?
It is more checkable, which is not the same thing. A grounded answer can be traced to a page and corrected at the source, but it inherits whatever that page got wrong. An ungrounded answer may be right and still gives you nothing to verify.
Can I make an AI assistant cite my website?
You cannot make it, but you can make it likelier. Grounded answers lift short self-contained statements from pages that were open, so a page that states your price, service area and method plainly in single sentences gets quoted more than a page that implies them.
Why does the same question sometimes cite sources and sometimes not?
The decision to go and look is made per question and can change with wording, timing, and which tool you are in. Add a word like current or 2026 and the same assistant often switches to fetching, which is why one buyer sees links and another does not.
Does grounding mean the AI actually read my whole page?
Usually not. It typically pulls the part of the page that matched the question and works from that. This is why one clear paragraph answering a real buyer question outperforms a long page where the same answer is spread across four sections.
How do I find out whether AI cites me or just describes me?
Ask the same buyer question in one tool that shows sources and one that usually answers directly, and compare. If you are named in neither, you have a corroboration problem. If you are cited but described wrongly, you have a page problem.
What does the AnswerHalo free check include?
It asks ChatGPT 25 real buyer questions from your niche and emails you a plain-English report: your score, who got named in your place, and the one fix to start with. It costs $0 and it runs as soon as you ask for it.
Your buyers are already asking. Find out what AI tells them.
The free check asks ChatGPT 25 real buyer questions about your niche and sends you the report within the hour, at $0. If it shows you are already getting named everywhere, we will say so in plain words.
Get my free check →