I have spent two decades watching marketers optimize the wrong unit. They optimize pages. They optimize posts. They optimize campaigns. Meanwhile, the systems that increasingly decide who gets discovered, cited, and recommended have quietly moved to a different unit entirely: the entity. The brand as a named, disambiguated, machine legible thing.
Here is the uncomfortable part. When a language model decides whether to mention your company in an answer, it is not evaluating your latest blog post. It is consulting an accumulated impression of you, built from every mention, every structured record, every consistent or inconsistent fact it has ever absorbed about your name. That impression behaves less like a score and more like mass. It builds slowly, it attracts more of what it already has, and once it reaches a certain density it starts bending discovery toward you without you doing anything on the day of the query.
I call this Entity Gravity: the accumulated, machine held weight of an entity that pulls citations, recommendations, and retrieval toward it in proportion to its mass. It is the single most undervalued asset in marketing right now, and my prediction is blunt. By 2029, entity gravity will explain more variance in AI driven discovery than content volume, domain authority, and ad spend combined.
Key Findings
- Machine trust attaches to entities, not URLs. A page inherits credibility from the entity behind it far more than the entity inherits credibility from any single page.
- Entity gravity compounds through consistency. Identical facts repeated across independent surfaces add mass. Contradictory facts subtract it faster than silence does.
- Gravity is query independent. A brand with high entity gravity gets pulled into answers it never optimized for, which is where most of the upside hides.
- Volume without coherence adds noise, not mass. A thousand thin mentions of an ambiguous name can weigh less than fifty precise ones.
- Small brands can outweigh large ones inside narrow topical orbits, because gravity is measured per context, not globally.
- The accumulation window is open now and will narrow. Entities that establish mass before AI systems mature will enjoy an incumbency effect that late arrivals pay dearly to overcome.
- Audit your entity before your content. Search your brand and founders across models and engines, note every wrong fact, every ambiguity, every split identity, and fix those before commissioning another article.
- Write your canonical fact set: legal name, founding year, leadership, locations, product names, one sentence description. Enforce it everywhere with the fanaticism of a legal department, because to a machine, variance reads as doubt.
- Choose two or three topical orbits where you can plausibly become the densest entity within eighteen months, and concentrate every mention, partnership, and byline there. Refuse gravity you cannot hold.
- Treat every rebrand, domain change, or renaming as an entity migration project with an owner, redirects, updated structured data, and third party corrections. Assume it takes a year, not a sprint.
- Add a standing quarterly review: ask the major assistants who leads your category and what they know about you. Track your presence in those completions the way you once tracked rankings. That number is your gravity, and from here on, it compounds for whoever starts first.
Trust Moved From Documents to Things
Classic search trained us to think in documents. Ten blue links, each judged on its own merits, each competing on relevance and links. That model is dying, and I wrote about the endgame in my piece on the future of Google search. What replaces it is a model where the system first resolves what things a query is about, then decides which entities it trusts on that subject, and only then selects documents as evidence.
This inversion changes everything about how credibility works. In the document era, a brilliant article from an unknown source could rank. In the entity era, the unknown source is the problem. The system asks: who is saying this, do I recognize them, does what I know about them support letting them speak on this topic? If the answer is a shrug, the content barely matters.
Think of it as trust mass: the density of confirmed, consistent, machine readable facts attached to your name. Trust mass is what the model reaches for when it has to complete the sentence "the leading firms in this space include..." with no page in front of it. You are either in that completion or you are not, and no landing page fixes it at query time.
The Physics of Accumulation
Gravity metaphors fail when they stay poetic, so let me make this operational. Entity gravity accumulates through four mechanisms, and I rank them by leverage.
First, identity resolution. The machine has to know that Pierre Subeh the author, Pierre Subeh the CEO, and Pierre Subeh in a podcast transcript are the same node. Every ambiguity splits your mass across duplicate entities, and split mass is lost mass. This is why I treat knowledge panels and entity SEO as infrastructure, not vanity.
Second, factual consistency. Same founding year everywhere. Same role titles everywhere. Same product names, spelled the same way, everywhere. Machines score agreement across independent sources, and agreement is the cheapest mass you will ever buy.
Third, contextual association. Gravity is topical. A hookah hardware brand accumulates mass in its category orbit, not in fintech. Repeated co occurrence with the right topics, the right competitors, and the right vocabulary tells the system where your mass belongs.
Fourth, temporal persistence. An entity that has said stable things for eight years outweighs one that appeared eight months ago, even at equal volume. Age is not a ranking factor here, it is a physics property. Longer observation windows produce tighter confidence.
Why Gravity Beats Volume
Marketers hear this and reach for the old playbook: publish more. That instinct is wrong, and it is wrong in a way that costs real money. Publishing a hundred mediocre pages under a weak entity is like throwing gravel into orbit and hoping it becomes a planet. It does not accrete, because nothing binds it.
The binding agent is coherence. Ten pieces that state the same core facts, attributed to the same resolved entity, connected to the same topical neighborhood, will add more mass than a hundred pieces that scatter. This is also why acquisitions and rebrands are so dangerous in the machine era. Change your name carelessly and you do not transfer your gravity, you orphan it. I have watched companies spend seven figures on a rebrand and walk away from a decade of accumulated machine trust because nobody in the room owned the entity migration.
There is a threshold effect here worth naming. Below a certain mass, you are effectively invisible to generative systems: not distrusted, just unresolved. Above it, you become a default completion, the name the model reaches for when a category comes up. The transition between those states is nonlinear, which means the marginal value of consistency work is highest exactly when it feels most pointless, in the early invisible phase where nothing seems to move.
Gravity Is Rented in Context, Owned in Aggregate
One nuance most analyses miss: entity gravity is not one number. It is a field, measured differently in every topical context. A regional accounting firm can have more pull than a global consultancy inside the query space of one island's tax code. This is the best news in this paper for small brands.
The strategic implication is to pick orbits you can dominate. Suppose, back of the napkin, that establishing meaningful gravity in a narrow niche takes fifty coherent, corroborated surfaces, while a broad category takes five thousand. The niche player who owns their fifty becomes the machine's default answer for an entire query neighborhood, and defaults are sticky. Broad players will find, painfully, that their global reputation does not automatically translate into local pull, because the machine never saw them do the local work.
Citations follow the same field logic, and the tactical layer of earning them is a discipline of its own that I covered in my work on LLM citation optimization. But citations are the harvest. Gravity is the soil.
The Incumbency Window
Here is my falsifiable claim. Between 2026 and 2030, AI systems will increasingly cache entity level trust rather than recompute it, for the same reason search engines cached link authority: it is expensive to rebuild and stable enough to reuse. Cached trust means incumbency. The entities that are dense, consistent, and well resolved when that caching hardens will enjoy a structural advantage that later entrants cannot buy with spend, only with years.
If I am right, we will see it in the data by 2028: model recommendations in established categories will show measurably lower churn than the underlying market share, brands will complain that AI systems describe them as they were three years ago, and a correction industry will emerge. If model outputs stay perfectly fluid and newcomers displace incumbents in machine answers as fast as they do in the market, I am wrong, and you can hold me to that.
I doubt I am wrong. Every information system I have ever optimized against, from Google's index to the App Store to social graphs, developed inertia as it matured. Trust systems calcify. That is not a flaw, it is what trust is.