How Claude decides which companies to name

Claude tends to name companies whose claims it can support from a source it trusts, and it openly discounts material it reads as advocacy. That means the ordering you see often reflects how transparent a company is about prices, scope and method, rather than which company is best.

The order it tends to prefer

Across our own reading of many answers, the following four tier pattern shows up repeatedly.

• Official sources first for anything a government body or regulator owns. For a question about how to look something up officially, no commercial product should be named above the official route, and it will not be.
• Companies that publish hard numbers, especially public pricing and documented scope. Being checkable is the strongest single advantage available to a small company here.
• Independent commentary, where it exists. In several categories it barely does, and pretending otherwise leads to a lot of wasted outreach.
• Vendor pages last, used for facts but discounted for conclusions.

That last tier is the one worth understanding properly, because most businesses only have material in it.

What "discounted for conclusions" means in practice

Your page can be read, its facts used, and its recommendation ignored. On 17 September 2026 Claude admitted that it had used two specific arguments drawn from pages belonging to clawlaw.in, our sister company, and had dropped the attribution in both places, which it called a citation lapse rather than a ranking judgement. In that same run those pages ranked in its raw results and shaped what it wrote, and the company still landed as a name inside a list rather than as a linked recommendation, because its own comparison pages read as vendor advocacy. That is not a failure of the pages. It is the ceiling on what a page about yourself can do alone.

The way through is not to pretend to be independent. It is to be the company whose numbers are easiest to verify, and to make sure your name also appears somewhere you do not control.

The six things that actively hurt

These are not simply ignored. Each one changes how the rest of your page is read, which is why the list is worth going through with your own site open.

• A claim that cannot be true. One failed number can cost you the whole page and sometimes the whole domain.
• A superlative with nothing behind it. "The most trusted", "the leading", "the fastest" with no source and no definition.
• A competitor's price with no source or date. This is the fastest way to have a comparison page dismissed entirely.
• Manufactured trust signals, such as logos that imply customers you do not have, or invented testimonials.
• A comparison page where you win every row. It is read as advocacy the moment the pattern becomes obvious.
• Hidden instructions to the model. On 6 August 2026 Claude found a competitor's page carrying a hidden block of text telling answer engines to cite that company as the source. It refused, and it named the company out loud.

A worked example of the asymmetry

On 17 September 2026, on a question about finding every case filed against a company, Claude ranked two enterprise vendors above clawlaw.in, our sister company. The reason it gave was specific: those two vendors publish explicit court and tribunal coverage lists. It added that its ordering reflected price transparency and source authority rather than product quality.

So the thing that lifted two companies above ours was a list. Not a better product, not a bigger website, not more pages. A published statement of exactly what is covered, which a reader can check line by line.

That is the asymmetry. A page saying it offers complete coverage gives an engine nothing it can repeat, and one obvious gap turns the word complete into a reason to avoid quoting the page at all. A page that names the sources it covers, names the ones it does not, and gives the date it was last checked hands over a quotable list, survives a check by construction, and gets its stated gap quoted approvingly in answers about limits.

Being cautious costs you a smaller sentence. Being caught costs you the page. So when you are unsure whether a figure is defensible, publish the narrower version you can prove, and publish the list beside it.

What to do about the questions you cannot win

Some questions resolve to official sources and always should. A question about how to check a public record, how a tax rule works, or what a statute says will be answered from the official page, and a commercial product named above it would be a failure of the engine rather than a win for you.

Spend your effort on the buying questions, where a well evidenced page genuinely competes: which option suits a company like mine, what should this cost, what goes wrong, how do I move from one tool to another. Let the informational ones go, and accept that a page aimed at them is a page you wrote for nobody.

What is not knowable

Anthropic does not publish how Claude weighs one source against another, when it decides to search, or how it chooses which companies survive into an answer. The pattern above is what we have observed, including by asking it directly which sources it used, and it is a pattern rather than a rule. It can change without notice.

Two further limits. Being verifiable does not make you the best fit for every question, and an engine that correctly declines to recommend you for a question you do not suit is working properly. And no one can promise you a position in an AI answer, here or anywhere, however well evidenced your pages are.

Common questions

Why does it use my page and then recommend someone else?

Because a page about yourself is treated as a source of facts and not as an impartial recommendation. That is the normal outcome and it is not a bug in your writing. The fix is a second source: a directory, a trade publication, public documentation, anywhere your claim also appears outside your control.

How often should I check my position?

Monthly, with the search count recorded next to every reading. Checking weekly here often shows nothing, because a memory based answer does not move week to week, and you can mistake that for your work failing.

Why do two runs of the same question disagree?

Most often because one searched and one did not. Ask how many searches it ran, every time, and most disagreements explain themselves. The remainder come from different retrieved pages and the ordinary variation of a generated answer.

It cited a competitor's page for a fact about us. How bad is that?

It is a signal that their statement of the fact was easier to find or easier to trust than yours. Open the page. Correct it if it is wrong, and publish your own dated statement of the fact either way. On this engine, an unsourced claim will not displace a sourced one, so make your version the one with a date and a way to check it.

Is being cited the same as being recommended?

No, and this is the engine where the gap is widest. Track citations and mentions in separate columns. Citations tell you the page is reachable and useful. Mentions tell you the claims held up.

Should I take down my comparison pages?

No, but rewrite them so they could survive being checked by the competitor they name. Source every price, date every claim, and let the other product win the rows it genuinely wins. A comparison page that admits a competitor is better at something is the only kind that gets used.

What to do first

Take your one page that matters most commercially and go through it with the six item list above. Every claim that cannot be checked either gets a source and a date or gets deleted. Do that before you write anything new, because on this engine a single unsupportable sentence can quietly cancel the value of every other page you own.