How to Get Your Content Cited by AI: The Entity Positioning Signal Hiding in Your Sentences

You write a page that answers a question in your niche better than anything else on the first result screen, and then you ask ChatGPT or Google's AI the same question and it cites someone thinner than you. It is one of the most common and most frustrating experiences a Digital Publisher has right now, because the usual explanations do not fit. Your content is not thin, your facts are correct, your page is indexed. New research from Google points at a cause almost nobody is looking at, and it lives at the level of a single sentence. The order in which you name things, the subject and the object, changes whether an AI can pull your page into an answer at all.

Search Engine Journal reported on the finding on August 17, 2026, and it traces back to a Google research paper with a memorable title, "Empty Shelves or Lost Keys? Recall Is the Bottleneck for Parametric Factuality." The short version is that AI models already know far more than they can reliably retrieve, and the way a fact was written, specifically which entity came first, decides whether the model can get to it. For a publisher trying to earn AI citations, that turns a vague goal into a concrete writing habit. PIB calls that habit Entity Positioning, and it is one of the most testable moves in the whole GEO toolkit.

What Google actually found: the model knows it but cannot recall it

Start with the surprise in the paper, because it reframes the problem. When Google tested frontier models such as Gemini-3-Pro and GPT-5, it found that they had encoded between 95 and 98 percent of the facts in the test set. The knowledge was in there. Yet on direct questions the same models failed to recall 26 to 34 percent of those encoded facts, and even when the models were allowed extended thinking time, 11 to 12 percent stayed out of reach. The bottleneck was not missing training data. It was access to knowledge the model already held.

The researchers named a specific trigger for that recall failure, and this is the part every writer needs. Models struggle to recall a fact when the question reverses the subject and object order from the way the fact appeared in the source text. Google's own example uses the sentence "Oasis played their first gig at the Boardwalk club." That sentence puts Oasis in the subject position and the Boardwalk club in the object position. Ask the model in the same direction and it answers. Flip the direction of the query so the object comes first, and the model stumbles, even though it clearly holds the fact. Put the same reversed pair into a multiple choice question and the model recognizes the right answer immediately, which proves the information was encoded and simply could not be recalled on demand.

Read plainly, an AI model does not store a clean fact you can query from any angle. It stores a pattern in a particular order, and the order you wrote it in becomes part of how retrievable that fact is later. That is a strange idea if you think of these systems as databases, and an obvious one if you think of them as language machines that learned from the shape of sentences.

Why sentence order decides who gets cited, not just who gets recalled

Recall inside the model is one half of the picture. The half that pays a publisher is citation, the moment an AI answer names your page as the source. A separate 2026 study by Advanced Web Ranking looked directly at that, comparing passages that AI engines quoted against passages the engines surfaced but did not cite, and it found the same signal from the other direction.

Passages the engines cited named the entity in the subject position at first mention 96 percent of the time. Passages the engines surfaced but did not cite named the entity in the subject position only 82 percent of the time. That gap of 14 points is large in a world where a citation is worth real traffic and a non-citation is worth nothing. The engine is not reading for beauty. It is looking for a passage it can lift cleanly, where the subject, the attribute, and the value sit in plain order, and it visibly prefers not to do the extra work of figuring out that "it" or "this company" means you.

The same study surfaced the other traits that separate cited passages from ignored ones, and they line up with how a retrieval system thinks. Cited passages tended to carry a visible recent date, roughly 80 percent of them, against 53 percent of the uncited group. Cited passages used an extractive structure, a clean claim a machine can quote without editing, about 76 percent of the time. Most striking, a hard number or a novel claim showed up almost only in the cited group, 6 percent of cited passages carrying a hard number and 7 percent a novel claim, against zero of the uncited passages on either count. A passage that only restates the consensus gets absorbed into the model's general understanding and earns no link. A passage that adds one specific, attributable fact gives the engine a reason to point at you.

Put the two studies together and they tell one story. Google's paper explains why order matters inside the model, and the citation study shows that the same preference decides which live page gets the link. Both point at a writing habit you control on the page, before any tool or schema enters the picture.

Entity Positioning, defined

Entity Positioning is how clearly and consistently your page names its key entities, in the subject position, at first mention, in the word order a reader would actually use to ask about them. It is one of the core measures inside GEO, the practice of Generative Engine Optimization, which is how Digital Publishers stay visible when an AI answer replaces the old list of blue links. Where Topical Authority is about the depth and breadth of what you cover, Entity Positioning is about the sentence-level clarity that lets an engine attach a fact to your brand and retrieve it later.

An entity is any named thing the answer is about, a company, a product, a person, a place, a method. The failure mode is writing around those names instead of leading with them. "It has become one of the more reliable ways to earn on Facebook this year" tells an engine almost nothing it can attribute. "The Facebook Content Monetization program pays publishers for qualifying views" names the entity, states the attribute, and gives the value, all in an order a machine can lift. Same fact, completely different retrievability.

A machine will not do the work of deciding that "it" means your brand. Name the entity, in the subject position, the first time and every time.

Five sentence-level moves that earn the citation

None of this requires new software. It is a set of habits you apply while writing and while editing, and you can test them, which is the part that matters most. Here is the short list, drawn straight from what the two studies reward.

Lead with the entity, not a pronoun or a "this"

Open each key passage by naming the thing the passage is about. Replace "This approach has become important lately" with "Entity Positioning determines whether an AI answer can attribute a fact to your page." The first version makes the engine hunt for the subject and often gives up. The second hands it a clean subject-attribute-value triple. This one change maps to the 96 percent versus 82 percent gap between cited and uncited passages in the citation data, which makes it the highest-return edit on the list.

Match your word order to how people ask

Because reversing subject and object hurts recall, write your load-bearing facts in the direction your audience phrases the question. If publishers ask "how much does Facebook Content Monetization pay," lead with the program as the subject and the payout as the object, in that direction, rather than burying the payout first and the program at the tail of the sentence. You are aligning the shape of your sentence with the shape of the query, which is exactly the alignment Google's paper says the model depends on.

Put one hard number in the passage

Zero uncited passages in the citation study carried a hard number or a novel claim. A specific figure, a real payout, a measured result, a dated statistic, gives an engine something concrete to quote and a reason to attribute it to you rather than absorb it silently. A number from your own operation is stronger still, because no other page has it. Vague ranges do the opposite and read as filler.

Add one claim that is not the consensus

If your passage only restates what every other page says, the model already holds that consensus and does not need to cite anyone for it. A distinctive claim, an angle from having run the play, a correction of a common mistake, a result that cuts against the received wisdom, is the thing an engine points at because it cannot get it anywhere else. This is where operator experience turns directly into citations.

Keep the entity neighbor correct

The citation study found that a passage defining the wrong neighboring concept lost its citations entirely. If you are trying to own "Entity Positioning," do not let the passage drift into defining "topical relevance" instead and hope the engine sorts it out. Name the exact entity you want to be cited for, define that one, and keep the surrounding sentences pointed at the same subject so the engine's picture of what your page is about stays sharp.

Where Entity Positioning fits in the larger GEO picture

Entity Positioning is one signal, and it works because it feeds the bigger measure that actually moves revenue, AI Citation Presence, the share of AI answers in your category that name your brand as a source. Clean sentence-level positioning is what lets an engine trust and attribute a source, and it compounds with the other GEO signals rather than replacing them. Topical Authority gives the engine a reason to consider you an expert on the subject. Technical Retrievability, the schema and structure work that makes your pages machine-readable, removes the friction on the crawl side. Entity Positioning is the writing layer that ties a specific, quotable fact to your name. We cover how these fit together for a publishing business in our guide to GEO for publishers in 2026, and how to watch the traffic it produces in AI search traffic in Google Search Console.

The reason this belongs at the center of a publisher's plan, rather than the edge, is that AI answers are becoming a primary way readers find sources at all. When an engine answers the question and names three publishers, being one of the three is the difference between visibility and silence. That is also why it cannot be the only channel you trust. A publishing business that earns across Facebook, Google Discover, content syndication, AI search, and asset value is far harder to shake when any single surface changes its rules, which is the entire point of running a publisher operating system rather than depending on one feed. Entity Positioning is how you compete for the AI search slice of that mix, and it costs you nothing but attention to your own sentences.

The method: this is analysis and optimization, not a one-time rewrite

The habit above will lift your odds. The result comes from treating it as a loop, which is the actual work PIB delivers and the opposite of rewriting a page once and walking away. You write for Entity Positioning, then you check whether it worked, then you push more of what earned the citation.

Checking is the step most publishers skip, and it is not hard. You ask the AI engines the real questions in your category, across ChatGPT, Google's AI answers, Gemini, and Perplexity, and you record which pages get cited and which entities the engines associate with your brand. You watch your own analytics for referral visits arriving from the AI assistants, because Google folds AI feature traffic into the aggregate Web search type in Search Console rather than breaking it out as a separate row. Where you want the machine-readable layer to match the writing layer, you mark up entities with schema, Organization and Article types with real dates and a clean sameAs graph, so the structural signal agrees with the sentence you wrote. Then you read what is already earning the citation and you write more of it, the same way the two lead pillars of any page work, Curation, what you choose to publish and how you shape it to your audience, and Virality, the reach that turns one strong post into a monetized event. A watermark can mark your words, but it cannot mark your read on your own data, and neither can a competitor copy it.

If you want that loop run as a system rather than assembled by hand, The GEO Authority System is $499 and packages the LLM Visibility Evaluation, the GEO Authority Playbook, and the distribution flow that puts these signals to work across your pages. If you would rather have experts train your team so you keep 100 percent of the upside, that is Consulting, and if you want the whole publishing operation, including AI visibility, run for you on a revenue share with no money upfront, that is Turnkey Management. The writing habit is free and you can start today. The system exists for when you want the measurement, the schema, and the optimization loop handled end to end.

Frequently asked questions

What is Entity Positioning in simple terms?

Entity Positioning is naming the key things your page is about, clearly and in the subject position, the first time you mention them, in the word order a reader would use to ask about them. It is a sentence-level writing habit that makes it easy for an AI engine to attach a fact to your brand and quote your page, rather than absorbing the information without crediting you. It is one of the core signals inside GEO, alongside Topical Authority and Technical Retrievability.

How does word order actually change whether AI cites me?

Google's 2026 research found that frontier models fail to recall 26 to 34 percent of facts they have encoded, and that reversing the subject and object order from the source text is a specific trigger for that failure. A separate citation study found that passages the engines cited named the entity in the subject position at first mention 96 percent of the time, versus 82 percent of the passages the engines surfaced but did not cite. Writing your load-bearing facts in the direction people ask questions aligns your sentences with how the model retrieves, which raises your odds of being the cited source.

Do I need schema or technical work to benefit from this?

No, the writing habit stands on its own and is the highest-return place to start, because it costs nothing but attention to your sentences. Schema and structured data help by making the machine-readable layer agree with the writing layer, which is Technical Retrievability, and they matter more as you scale. Start by fixing how your passages name their entities, then add the structural markup so the two signals reinforce each other.

How do I know if it is working?

Ask the AI engines the real questions in your category, across ChatGPT, Google's AI answers, Gemini, and Perplexity, and record which pages get cited and which entities the engines link to your brand. Watch your analytics for referral visits arriving from the AI assistants, because Google Search Console folds AI feature traffic into the aggregate Web search type rather than reporting it as its own row. Treat it as a repeating check rather than a one-time audit, because the answers shift as engines update and as your content library grows.

Is this the same as old-school SEO keyword placement?

No, and the difference matters. Keyword placement was about matching strings a ranking algorithm counted. Entity Positioning is about giving a language model a clean, attributable subject-attribute-value statement it can quote and trace back to you. The goal is not density, it is retrievability and attribution, which is why one distinctive number or claim in a passage outperforms repeating the same phrase many times.

Does this replace writing useful content?

No, it makes useful content findable. The citation studies are clear that thin, consensus-only passages get absorbed and earn no link no matter how they are ordered, while passages carrying a real number or an original claim get cited. Entity Positioning decides whether an engine can quote the value you already created. The value still has to be there first, and first-hand operator experience is what the engines reward.

Key takeaways

• Google's 2026 paper found frontier models encode 95 to 98 percent of tested facts but fail to recall 26 to 34 percent on direct questions, and reversing the subject and object order from the source text is a specific trigger for that recall failure.
• A separate 2026 citation study found passages the engines cited named the entity in the subject position at first mention 96 percent of the time, versus 82 percent of the passages the engines surfaced but did not cite.
• Entity Positioning is the writing habit that captures both findings, naming your key entities clearly, in the subject position, at first mention, in the word order readers use to ask.
• Cited passages also tend to carry a visible recent date and an extractive quotable structure, and while a hard number or non-consensus claim was rare overall, it appeared only among cited passages, 6 percent with a hard number and 7 percent with a novel claim, versus zero of the uncited passages.
• Entity Positioning feeds AI Citation Presence and works alongside Topical Authority and Technical Retrievability, and it is the AI-search piece of a diversified publisher operating system.
• The result comes from a loop, write for Entity Positioning, test which pages the engines cite, then publish more of what earns the citation, which is analysis and optimization rather than a one-time rewrite.

Sources

• Search Engine Journal, Google: Subject/Object Entity Order Affects AI Answers (2026-08-17): https://www.searchenginejournal.com/google-subject-object-entity-order-affects-ai-answers/586089/
• Google Research, Empty Shelves or Lost Keys? Recall Is the Bottleneck for Parametric Factuality (2026): https://research.google/
• Advanced Web Ranking, What Gets Quoted and What Gets Absorbed: A Passage-Level Study of AI Citations (2026): https://www.advancedwebranking.com/blog/passages-quoted-vs-passages-absorbed-in-ai-answers
• Google Search Central, Understanding AI features in Search and Search Console reporting: https://developers.google.com/search/docs/appearance/ai-features
• Google Search Central, E-A-T gets an extra E for Experience (2022-12): https://developers.google.com/search/blog/2022/12/google-raters-guidelines-e-e-a-t