Strings versus things
Type "SEO agency" and a search engine matches documents containing that string and its variants. Type "Last Agency" and it should be doing something different: retrieving a record about an organisation — founders, city, category, services, prices, the other sites that mention it — and answering from that record.
Google announced this shift in 2012 when it launched the Knowledge Graph, under the phrase "things, not strings". Everything since — Knowledge Panels, entity-based retrieval, and now the AI systems answering above the results — runs on the same idea. A query gets resolved to entities first, and documents get judged on how well they relate to those entities.
The practical consequence: optimising for a string means writing pages that match phrases. Optimising as an entity means making yourself a coherent, corroborated record that a machine can be confident about.
How Google decides you exist
Confidence comes from repetition across sources that don't control each other. Your own website is a single, self-interested source. It states that you're an SEO agency in Bangalore founded in a given year. That's a claim. When a trade publication, a directory, a podcast page, a conference speaker bio and a customer review all describe the same organisation the same way, it stops being a claim.
This is why a well-funded startup with a beautiful website and no press has no Knowledge Panel, and a dull thirty-year-old manufacturer with a Wikipedia stub and forty trade mentions has one. Wikipedia and Wikidata are heavily used sources — not because they're magic, but because they're structured, edited and cited.
It's also why the answer to "how do we get a Knowledge Panel" is never a schema tweak. See knowledge graph for what the record actually holds.
The four signals you control
You can't grant yourself entity status. You can make it much easier to grant.
- Organization schema with `sameAs`. Mark up your organisation once, sitewide, with the legal name, logo, address, contact points, founding date and founders. Then list every profile you control in
sameAs— LinkedIn, Crunchbase, YouTube, Instagram, X, your GitHub if you have one.sameAsis how you say "these accounts and this company are the same thing". More on the implementation in schema markup. - One name, spelled one way, everywhere. "Last Agency", "LastAgency", "Last Agency Pvt Ltd" and "lastagencyhere" are four strings competing to be you. Pick the public-facing name and make every profile, invoice footer, directory listing and email signature match it, capitalisation included.
- Third-party mentions that repeat the same facts. Founder name, city, category, what you sell. Directory profiles, podcast bios, speaker pages, press coverage, partner pages. The facts matter more than the links here — an unlinked mention still corroborates.
- A page that states the facts plainly. An About page with legal name, registered address, founding year, named founders and a clear description of what you sell. It's boring to write and it's the page most often quoted when an assistant summarises a company, because it's the only one that answers directly.
Why this matters more when a model answers the question
Ten-blue-links search let a strong page carry a weak brand. If your article matched the query best, you ranked, whether or not anyone knew who wrote it. Generative answers work differently: the system resolves the question to entities, retrieves from sources it has some confidence in, and names the ones it can attribute.
If your company isn't a resolvable thing, you're only quotable when a specific page happens to match — never when the question is "who should I hire in Pune" or "which vendors do this". Being nameable is now a distribution channel, which is a strange sentence and an accurate one.
The failure mode to worry about is contradiction, not absence. If three sources give your company three founding years and two different cities, a model won't hedge — it'll pick one and state it flatly. Fixing conflicting facts across your own profiles is cheaper than any content programme, and it's the first thing we check when a client asks why AI assistants describe them wrongly. Getting cited by AI assistants starts here.