Last spring I ran the same 30 prompts against three assistants for a client in a niche B2B category. Their competitor, roughly eight times their size with a domain twenty years older, appeared in 21 of the answers. My client appeared in two.
Six months later my client appeared in 19. Nothing about their domain authority changed meaningfully. What changed was that they became easy to corroborate.
That is the mechanism most people miss. Getting cited is not about being the biggest. It is about being the source that is cheapest for the system to trust.
Citation is a corroboration problem, not an authority problem
When a language model composes an answer, it is under pressure to not be wrong. The cheapest way to reduce that risk is to prefer claims that appear in more than one independent place and to prefer sources whose identity it can resolve confidently.
So the practical question is not "am I authoritative." It is: if a system encounters a claim on my site, how hard is it to verify that I am a real entity who plausibly knows this, and how hard is it to find the same claim somewhere else?
Everything below reduces that cost.
Become a resolvable entity first
Before content tactics, do the identity work. A model that cannot tell whether "X Network" is an agency, a telecom, or a social platform will not risk attributing anything to you.
Concretely: consistent naming everywhere, no abbreviation drift. Organization and Person schema with sameAs links pointing to the profiles that actually exist. A Wikidata item if you legitimately qualify. Author bios that state credentials in plain sentences rather than in a designed graphic, because an image of your credentials is invisible to retrieval.
When I did the personal SEO work on owning my own name, the side effect I did not anticipate was how much easier it became to get quoted, because there was suddenly one unambiguous answer to the question of who I am. The same principle scales to companies. Entity SEO and knowledge panels is the long version of this argument.
Run your key pages through a schema markup generator if you are not comfortable writing JSON-LD by hand, then validate it. Half the schema I audit on client sites is present but malformed, which is worse than absent because it signals carelessness.
Publish claims that only you can make
Models heavily favor content that adds information not already in their training. If your page restates the consensus, there is no reason to cite you over the ten thousand other pages that say the same thing.
The kinds of claims that are structurally uncitable elsewhere:
Original counts. "We reviewed 140 checkout flows in the Caribbean market and 61 of them required account creation before showing shipping cost." Nobody else has that number. You do, because you counted.
Process specifics. Exact sequences, exact thresholds, exact tooling. "We migrate in this order: redirects, then internal links, then sitemaps, then GSC property, and we wait 14 days between the last two."
Named tradeoffs. "This approach fails when your catalog exceeds roughly 5,000 SKUs, because the crawl budget math stops working."
Firsthand negatives. What did not work is rarer than what did, and rarity is citation fuel.
I do not fabricate a study to get this. I count things I already have access to and report the count honestly, including the sample size and the limitation. That honesty is itself a trust signal, and it is the difference between real experience and the manufactured statistics that saturate this space.
Get corroborated off your own domain
A claim that lives only on your site is a claim with one witness. Three moves I use, in rough order of return.
Original data plus outreach. Do a small original count, then pitch it to publications that cover your category. When the number gets repeated on three sites you do not own, it stops being your marketing claim and becomes a fact about the world with your name attached. This is the actual point of digital PR done properly, and it is why link buying never produced this effect.
Podcast appearances with transcripts. Spoken statements attributed to a named person, published as text, are excellent corroboration. Prioritize shows that publish full transcripts over shows with bigger audiences and no text. I go deeper on the selection criteria in podcast guesting as a growth channel.
Answers in communities that get indexed. Detailed, non promotional answers under a real identity on forums that rank well are retrieved constantly. I have written separately about why forum threads outrank brand pages, and the same dynamic applies to citation.
Structure so a passage can be lifted cleanly
Even a perfect claim gets skipped if it is buried in a paragraph that only makes sense in context.
Every citable passage should survive being cut out with scissors. Under a heading that states the question. Answer in the first sentence. Self contained, meaning no "as noted above," no unresolved pronouns, no reliance on the previous paragraph.
I run a crude test on drafts: I delete the article title and every heading, shuffle the paragraphs, and see which ones still make sense alone. The ones that survive are the ones that get quoted. Usually about a fifth of a first draft passes. Getting that to half is the entire editing job.
What I tell clients not to bother with
Stuffing an llms.txt file with marketing copy. The file is worth publishing as a clean map of your important content, and it takes about ten minutes, but it is a courtesy signal, not a ranking lever. Anyone promising you citations from it is selling something.
Rewriting everything into FAQ format. Overdone FAQ blocks read as spam to humans and dilute your passage quality. Use them where there are genuinely discrete questions.
Chasing mentions in every category. Pick the six to ten questions where a mention actually leads to revenue and go deep. Being cited on "what is marketing" is worth nothing.
Obsessing over daily fluctuation. Answers vary by session and phrasing. I test a fixed prompt set monthly, same wording, same account, and I only act on changes that hold for two consecutive months.
The uncomfortable part
This work has a long lag. The client I opened with saw almost nothing for four months, then moved fast. Corroboration compounds because each new independent mention makes the next one easier to earn, which is the same compounding curve I described in why SEO behaves like compound interest.
If you need results this quarter, buy ads. If you want to be the default answer in your category three years from now, start counting something nobody else has counted, publish the number with its limitations stated plainly, and spend the next six months making sure it gets repeated by people who do not work for you. That is the whole game, and it is available to companies of any size, which is precisely why the incumbents are nervous.