A comparison question needs comparative sentences to answer it, and vendor sites do not credibly produce those about competitors. So engines answering “X versus Y” lean on third-party pages, and your competitor’s alternatives page can carry more weight than your homepage.
That is a sourcing problem rather than a content-quality problem, and it is why publishing more about yourself does not move these queries.
Comparison prompts are the highest-intent questions in this discipline. Somebody asking whether X or Y suits them has a budget and a shortlist, and the answer they get is assembled from pages neither vendor controls.
What does a comparison answer need?
A sentence that compares. That sounds trivial and it is the entire mechanism.
Anthropic’s citations documentation describes the feature as returning “the exact passages that support each claim, so you can verify answers and surface sources.” For a claim of the form “X suits smaller teams and Y suits larger ones,” the supporting passage has to be a passage that says something like that.
Your product pages do not contain such a passage. They contain claims about you, in isolation, which support a description and not a comparison.
Support an answer describing you. Cannot support an answer ranking you against a named alternative, because no sentence on them does that.
Support exactly the sentence a comparison question needs. This is why they win these queries almost by default.
Why does your own comparison page underperform?
It has the right shape and the wrong author, and only the second is a problem you can partly solve.
A vendor page comparing itself to a competitor does contain comparative sentences, so it is retrievable for these questions and often is retrieved. What it lacks is independence, and an engine assembling an answer from several sources will tend to have more of them agreeing with the third-party version than with yours.
There is also a coverage asymmetry that catches people out. Your comparison pages cover the competitors you chose to name. Third-party pages cover whoever the author chose, which is frequently a different set, and that difference is what puts you in the wrong company, covered in ai comparing us to wrong competitors.
Your versus page is a legitimate source and a single one. The answer is assembled from several, and you wrote one of them.
What makes a vendor comparison page actually usable?
Being genuinely useful to somebody deciding, which is also what makes it hard to write.
| Page property | Why it matters for retrieval |
|---|---|
| Names the competitor plainly, in the text | A page that says “other tools” cannot be retrieved for a question naming one. |
| States where the competitor wins | Produces the comparative sentence with a constraint attached, which is the shape the question needs. |
| Gives dated, checkable figures | A price with a date is quotable. A claim about being better is not. |
| Answers the constrained version | “Better for teams under twenty” beats “better for teams”, because the constraint is what buyers actually type. |
The second row is where most vendor pages fail and where the gain is. A page that concedes nothing supports no comparative sentence except its own conclusion, which is exactly the sentence an engine has least reason to reuse.
Does any of this need special markup?
No, and it is worth saying because comparison pages attract a lot of schema advice. Google’s documentation on AI features states that “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary,” and that a page “must be indexed and eligible to be shown in Google Search with a snippet.”
So the work is editorial. Write the comparison, make it accurate, get it indexed. There is no markup that makes a page eligible for comparison answers that is not already eligible.
What about the comparison you are not in?
It is the more expensive problem and the less visible one. A buyer asking whether X or Y suits them never learns you exist, and nothing about that appears in any report you run, because you were not in the answer to be counted.
The only way to see it is to run the pairings you are absent from. Take your two closest competitors, ask an engine to compare them to each other without naming yourself, and read whether you get mentioned as a third option. Often you will not, and that silence is the finding.
This is why a comparison audit should include prompts that do not contain your name. A set built only from your own pairings measures the questions you already compete in, and misses the ones where the market has settled on two names and you are not one of them.
How do you find out who is winning your comparison queries?
Microsoft’s AI Performance report in Bing Webmaster Tools helps on the other side of the same question, since it shows “citation counts for specific URLs from your site” and the grounding queries behind them. If your own comparison page is being cited, you will see it there.
What do you do about the pages you cannot control?
Three moves, in descending order of yield and ascending order of effort.
Get added where you are absent. A roundup covering your real competitors that omits you is the highest-value correction available, and it is a request rather than a project.
Correct what is wrong. A comparison page stating a price you no longer charge or a feature you now ship is a factual error most publishers will fix, and the ask is stronger when you point at your own public page as the source.
Publish the comparison nobody has written. If your genuine advantage is on a constraint no existing page addresses, the page answering that question has no competition, because no third party has bothered to write it. That is slower than the first two and it is the only one that compounds.
What does not work is asking to be moved up a ranking on somebody else’s page. It reads as exactly what it is, and the credibility of the page is the reason it carries weight in the first place. The broader version of that discipline is in third party sources ai citations audit, and the prompt set these comparison questions belong in is specified in how to build a prompt set for ai visibility tracking.