Entity SEO and Citations: How AI Engines Pick Who to Recommend

The short answer

Entity SEO is the work of making your company a clearly defined, consistently described entity that AI engines can resolve and trust. LLMs recommend vendors by retrieving what independent sources say about an entity in a category, not by ranking keywords. You earn LLM citations by owning one category, corroborating it across third-party sources, and publishing quotable proof.

Most B2B tech companies still think about search as a page problem: pick a keyword, write a page, earn links, climb the ranking. That model explains Google's ten blue links reasonably well. It explains almost nothing about why ChatGPT names three DevOps consultancies and not a fourth, or why Perplexity cites a competitor's Clutch profile instead of your homepage.

AI engines think in entities. When a CTO asks for the best data engineering partner for a HIPAA-regulated SaaS, the model is not matching a string. It is resolving a set of named things (companies, people, products, categories) and pulling the passages that describe how those things relate. If your company is a fuzzy entity, described differently on every platform and attached to no clear category, the model cannot place you and will not name you.

This guide is for the software and tech companies we work with at XQL: B2B SaaS teams, custom software and IT outsourcing firms, DevOps and data shops, and the Salesforce and HubSpot consultancies selling to technical buyers. It explains what an entity is to an LLM, how each answer engine chooses what to cite, and the specific entity SEO work that moves a B2B company from invisible to recommended. If you want the wider primer first, our AI search optimization guide for B2B tech covers AEO and GEO end to end.

What is entity SEO, and how is it different from keyword SEO?

Entity SEO optimizes how machines understand who you are, what you do, and what you are connected to. Keyword SEO optimizes how a page matches a query string. The first builds a stable identity a model can retrieve and trust. The second builds a page a crawler can rank. AI engines rely on the first far more than the second.

A keyword is text. An entity is a thing. Kubernetes is an entity. A specific DevOps consultancy is an entity. The relationship between the two, that the consultancy specializes in Kubernetes migrations for fintech, is a fact about entities that a knowledge graph can store and a language model can learn. When you write a page targeting the phrase Kubernetes consulting, you are hoping a crawler notices the string. When you build an entity, you are teaching every system that reads the web that your company and that category belong together.

Google has worked this way for over a decade. Its Knowledge Graph stores entities and their relationships, and it is what powers knowledge panels, disambiguation, and the increasingly semantic way queries are interpreted. LLMs learned the same relationships from the same web, but they hold them as statistical associations inside the model rather than as rows in a graph. Retrieval-augmented engines such as ChatGPT search, Perplexity, and Google AI Overviews then combine both: the model's internal sense of which entities belong together, plus live retrieval of the passages that describe them.

The practical difference shows up in the data. Ahrefs ran 15,000 long-tail queries through Google and Bing, then asked the same questions of ChatGPT, Gemini, Copilot, and Perplexity. Only about 12% of the URLs the AI assistants cited also ranked in Google's top 10 for the same prompt. Perplexity was the most Google-aligned at 28.6%, and the other three hovered around 8%. If citation followed keyword ranking, that overlap would be close to 100%. It is not, because the engines are choosing sources on a different basis.

How do AI engines resolve your brand into an entity?

AI engines resolve a brand by matching the name, the URL, and the descriptive facts about it across many sources until they converge on one identity with one set of attributes. The clearer and more repeated those facts are, the more confidently the engine can place you in a category and attach proof to your name.

Consistent naming is the entry ticket

Entity resolution starts with the name. If your site says Acme Software, your Clutch profile says Acme Software Solutions, LinkedIn says ACME, and Crunchbase lists Acme Technologies Inc., a machine has four candidate entities and weak evidence that they are one company. Human readers reconcile that instantly. Retrieval systems often do not. The same problem applies to your founders (one name, one title, one bio everywhere), your products, and your service lines. Pick the canonical form of every name and use it on every platform a machine might read.

Structured data tells machines what the name refers to

Organization schema on your homepage or about page is the cleanest way to declare who you are. Google's Search Central documentation states there are no required properties, and recommends including the fields that apply: url, logo, description, address, contact details, and sameAs. In November 2023 Google expanded the Organization properties it reads to populate knowledge panels and attribution, which is a signal of how much weight it places on this block. The sameAs array is the key entity feature. It lists the other URLs on the web that refer to the same organization, so a machine can merge your LinkedIn, Crunchbase, Clutch, G2, Wikidata, and GitHub presences into one node.

{
  "@context": "https://schema.org",
  "@type": "Organization",
  "@id": "https://www.example-software.com/#organization",
  "name": "Example Software",
  "legalName": "Example Software Inc.",
  "url": "https://www.example-software.com/",
  "logo": "https://www.example-software.com/logo.png",
  "description": "Example Software is a nearshore data engineering company that builds and maintains data platforms for B2B SaaS companies in regulated industries.",
  "foundingDate": "2015",
  "founder": {
    "@type": "Person",
    "name": "Jane Doe",
    "sameAs": "https://www.linkedin.com/in/janedoe/"
  },
  "address": {
    "@type": "PostalAddress",
    "addressLocality": "Krakow",
    "addressCountry": "PL"
  },
  "knowsAbout": ["Data engineering", "Snowflake", "dbt", "HIPAA compliance"],
  "sameAs": [
    "https://www.linkedin.com/company/example-software/",
    "https://clutch.co/profile/example-software",
    "https://www.crunchbase.com/organization/example-software",
    "https://www.g2.com/products/example-software",
    "https://www.wikidata.org/wiki/Q000000000",
    "https://github.com/example-software"
  ]
}

Two notes on that block. The @id gives your organization a stable identifier you can reference from every other schema object on the site, which keeps your Article, Service, and Person markup pointing at the same node. And the description should read exactly like the description on your LinkedIn page and your review profiles, because the point of structured data is not to say something new, it is to say the same thing in a form a machine cannot misread.

Corroboration is what makes the entity real

Schema declares. Third-party sources confirm. A model does not take your word for what you are, so entity resolution leans on the sources it already trusts to describe organizations. For a B2B tech company the corroboration set is fairly short, and each source plays a different role.

The order of operations matters. Fix the name and the description first, then push that identical description into the corroboration sources, then declare it in schema with sameAs pointing at those sources. Done in that order, every source a machine checks agrees with every other one, and the entity snaps into focus.

Why do third-party mentions and co-occurrence drive recommendations?

Third-party mentions drive recommendations because AI engines treat repeated independent description as evidence and treat self-description as a claim. Co-occurrence, meaning your name appearing near the category terms your buyers use, is how the model learns that you belong in the answer to that category question.

The strongest published evidence comes from Ahrefs, which studied 75,000 brands across ChatGPT, Google AI Mode, and AI Overviews. Brand web mentions correlated with AI Overviews visibility at 0.664, the highest of any signal in the original analysis, and the brands in the top quartile for web mentions earned up to ten times more AI Overviews mentions than the next quartile. An update to the same study found YouTube mentions the single strongest correlate across ChatGPT, AI Mode, and AI Overviews, at roughly 0.74. Backlinks, the currency of classic SEO, were far weaker. Ahrefs is careful to say correlation is not causation, and so are we, but the direction is consistent with how these systems retrieve.

Semrush's topic authority study, run with Kevin Indig across 1,094 US categories and more than 50,000 brands in ChatGPT between January and June 2026, adds the category angle. Only 15.2% of categories had a clear owner, defined as a brand that appeared in at least four of five related buyer prompts with a five-point lead. Another 31.2% had an emerging leader. The remaining 53.7% were unsettled, with no brand appearing in three of five prompts. Traditional SEO metrics barely predicted ownership: owners had higher organic traffic in only 48.4% of comparisons and higher Authority Score in 52.5%. The one metric that reached statistical significance was branded search volume, which is a proxy for how often people, and therefore the web, talk about you by name.

Read those two studies together and the mechanism is clear. A model recommends the entity that the web most consistently associates with the category. That association is built by co-occurrence: your name in the same passage as nearshore data engineering, or Salesforce for healthcare, or DevOps for fintech, across many independent sources. One mention is noise. The same association repeated across reviews, roundups, podcasts, community threads, and news is what the model treats as fact. And because more than half of ChatGPT categories have no owner yet, the association is still available to a focused firm that builds it deliberately.

Where the mentions live matters too. 5W's Citation Source Audit for Q1 2026 found Wikipedia (13.15%) and Reddit (11.97%) together account for more than a quarter of ChatGPT citations in the United States. For B2B tech, that does not mean chasing a Wikipedia page you will not get. It means the engines lean heavily on a small number of trusted reference and community sources, and the equivalent sources in your category, Clutch and G2 for reviews, the relevant subreddits and Hacker News for community, and credible industry roundups, carry the same kind of weight in the answers your buyers see.

How does citation selection work in ChatGPT, Perplexity, Google AI Overviews, and Gemini?

Each engine runs a version of the same loop: interpret the prompt, retrieve candidate passages, filter for trust and relevance, then synthesize an answer that names some entities and cites some URLs. The differences are in how much they retrieve, which index they lean on, and whether the brand they name is the source they cite.

ChatGPT search: broad retrieval, many citations, few named brands

When ChatGPT browses, it fans out into several sub-queries, retrieves a wide set of passages, and cites generously. Semrush's 2026 AI Visibility Index, built on 126 million US prompts, measured ChatGPT citing an average of 15 sources per response. Yet its answers name brands far less often than they cite them. Semrush's ghost citations study found ChatGPT cited a source 87% of the time but mentioned a brand in only 20.7% of answers. For a vendor prompt, that means ChatGPT reads a lot of the web about your category and then writes a short shortlist. Being one of the fifteen cited URLs is useful. Being one of the three names in the shortlist is the commercial outcome, and it depends on the entity work above rather than on any single page. Our guide to getting recommended by ChatGPT walks through that shortlist mechanic in detail.

Perplexity: the most search-like engine

Perplexity is a retrieval engine first. Every answer is grounded in live search results with numbered citations, and its source selection is the closest to conventional ranking. In the Ahrefs overlap study, 28.6% of Perplexity's cited URLs also sat in Google's top 10, more than three times the rate of ChatGPT, Gemini, or Copilot. So classic on-page quality, crawlability, and freshness still matter here, and a page that ranks well and answers directly has a real chance of being cited. Perplexity also weights recency and rewards passages that are explicit and self-contained, because it lifts them nearly verbatim. We cover the platform-specific tactics in how to rank in Perplexity for B2B.

Google AI Overviews: partly organic, increasingly its own thing

AI Overviews draw from Google's index and Knowledge Graph, so entity resolution here is the most literal. BrightEdge tracked AI Overview citations for sixteen months, from May 2024 to September 2025, and found that about 54% of cited URLs came from pages already ranking organically for the query, up from 32% at launch. That still leaves nearly half of citations going to pages outside the organic results. BrightEdge data also put AI Overviews on roughly 48% of tracked Google searches in early 2026, so for most B2B tech queries the AI answer now sits above the blue links. For a B2B tech company, this is the engine where a strong organic position, a resolved entity, and clean Organization schema compound. Our Google AI Overviews guide goes deeper on the Google-specific layer.

Gemini: names brands, cites sparingly

Gemini is the mirror image of ChatGPT. Semrush measured it citing only about three sources per response and found it named brands 83.7% of the time while citing a source in just 21.4% of answers. In practice, Gemini leans on the model's internal knowledge and Google's entity understanding, then names the entities it is confident about with little visible sourcing. That makes Gemini the purest test of your entity strength: if the model cannot describe you confidently from what it has already learned, it will not improvise a citation to cover the gap. It will simply name someone else.

Mentions versus citations: why 62% of citations never name the brand

A citation is your URL linked as a source. A mention is your company named in the answer. They are different signals, they diverge more often than most teams expect, and for a B2B company chasing pipeline the mention is the one that puts you in the deal.

Semrush's ghost citations study, run with Kevin Indig across 3,981 domains and four AI engines, found that 62% of observed AI citations were ghosts: the engine cited a page but did not mention the brand behind it. Only 21% of the most-cited domains in a category were also the most-mentioned brand, and the two measures correlated slightly negatively (-0.229). That result surprises people until they think about what gets cited. A comparison article on a publisher site earns the citation. The vendors inside the comparison earn the mentions. The publisher's domain authority wins the link; the vendor's entity strength wins the name.

This is the reason entity SEO and content SEO are separate strategies at XQL. Educational content, including this article, is built to be cited: a lead answer, one idea per paragraph, named sources, real numbers. Brand visibility is built to be mentioned: consistent positioning, corroborated category membership, and an authoritative roundup in your own category. The Princeton GEO paper, presented at KDD 2024, tested the content side directly and found that adding statistics, quotations, and source citations to a page lifted its visibility in generative answers by up to 40% on the position-adjusted word count metric. That is the citation lever. The mention lever is everything in the sections above.

What should a B2B tech company do about entity SEO?

Clean up the entity, position it in one category, corroborate that position across the sources engines trust, publish passages worth quoting, and declare it all in structured data. The six plays below are the order we run them for software and tech companies, because each one makes the next one land harder.

1. Run an entity audit and fix the inconsistencies

Start by asking the engines who you are. Prompt ChatGPT, Perplexity, and Gemini with your company name and read the descriptions back. Note every wrong fact, every stale service, every category the model attaches to you that you no longer want. Then inventory every place a machine can read about you: your site, LinkedIn, Crunchbase, Clutch, G2, Wikidata, GitHub, partner directories, press pages. Record the name, the one-line description, the category tags, the founder names, and the HQ on each. The spreadsheet will show you the fractures. A software development company that describes itself as a digital transformation partner on LinkedIn, a custom software firm on Clutch, and an AI consultancy on its own homepage has three weak entities instead of one strong one.

2. Position to one category and commit to it everywhere

The model recommends specialists because specialists produce clean associations. A firm described everywhere as the nearshore Salesforce partner for mid-market manufacturers becomes the obvious answer to that prompt. A firm that does Salesforce, HubSpot, custom development, and AI matches no prompt strongly. This is the hardest play commercially, because founders fear narrowing, but the Semrush data is on your side: more than half of ChatGPT categories have no owner, and ownership is decided at the topic level, across a cluster of related prompts, not on a single keyword. Pick the category you can win in the next two quarters, write the one-sentence description, and put that exact sentence on every profile from play one. You can expand outward once the first association is established. We built this play around exactly that sequence for Computools, a software development company we positioned as the recommended Salesforce partner inside the major LLMs.

3. Build the corroboration layer deliberately

Corroboration does not happen by accident for a B2B services firm. Nobody is writing about your company unprompted. So you build it. Claim or update the profiles in the corroboration table above and align every description. Ask recent clients for verified Clutch or G2 reviews and give them the category language, in their own words, so the review says Salesforce for healthcare rather than great team. Pitch inclusion in the credible industry roundups in your category and be honest about who else belongs on them. Get your founder or head of practice on the podcasts and YouTube channels your buyers watch, because Ahrefs found YouTube mentions the strongest correlate of AI visibility on every platform. Participate in the subreddits and communities where your buyers ask for recommendations, without spamming. The target is the same description of you, tied to the same category, in ten independent places.

4. Publish the authoritative roundup in your own category

Engines assemble shortlists from existing shortlists. A ranked, methodology-backed list of the best providers in your specific category, with fair and accurate descriptions of every competitor, is the single highest-leverage co-occurrence asset you can own, because it places your name next to the category term and next to the other entities the model already associates with it. It only works with integrity: state your evaluation criteria, describe others neutrally, and make your own case on evidence. Our roundup of the best B2B AEO agencies is the format applied to our own category.

5. Make your key pages quotable

Citations go to passages an engine can lift cleanly. Rewrite your service pages, comparison pages, and case studies so every section opens with a direct, self-contained answer in the first 40 to 60 words, under a heading phrased as the question a buyer would ask. Then support it with the things the Princeton GEO research found move generative visibility: a specific statistic, a named source, a quotation from a client or a named expert. A passage that says a Salesforce consultancy grew SQLs from organic 500% is quotable. A passage that says you deliver results-driven transformation is not, and no amount of schema will rescue it. Strip the adjectives and leave the numbers.

6. Declare it in structured data and keep it current

Add the Organization block with sameAs from the code example above to your homepage, and give it an @id every other schema object on the site references. Add Person markup for the founders and named experts with their own sameAs to LinkedIn. Add Service markup for each service line, and Article markup with a real author on every piece of content. None of this creates trust by itself, but it removes ambiguity, and ambiguity is what stops an engine from naming you. Then keep it current. AI engines penalize staleness in vendor categories because outdated information is a liability in the answer, so a founder who left, a service you dropped, or a five-year-old client logo all erode the entity.

How do you measure entity strength and LLM citations?

You measure it on three layers: how the engines describe your entity, how often you are mentioned and cited for the buyer prompts that matter, and what that visibility produces in branded search, AI-referred sessions, and CRM pipeline. Rankings dashboards do not show any of it, so you need a different instrument.

Semrush's AI Visibility Index found that 45% of marketing leaders cannot accurately measure brand visibility in AI answers and only 9% have tools that track every relevant metric. That gap is an advantage for the teams who measure at all. Here is the minimum viable measurement stack we run for B2B tech clients.

• Entity description check, monthly. Ask each engine to describe your company and to list what it is known for. Score accuracy, category match, and named proof. This is your entity health metric, and it should move first.
• Prompt panel, weekly. Take the 15 to 30 buyer prompts you want to win (best vendor for, top alternatives to, who should we hire for) and run them across ChatGPT, Perplexity, Gemini, and Google AI Mode. Record mention (named in the answer), citation (URL linked), position in the shortlist, and who else appears. Tools such as Semrush's AI Visibility Toolkit, Ahrefs Brand Radar, and Profound automate the volume; a manual pass on the ten prompts that drive revenue keeps you honest.
• Co-occurrence count, monthly. Count the independent pages where your name appears within the same passage as your target category term. This is the leading indicator for mentions, and it is the number your corroboration work should move.
• Branded search and direct traffic, monthly. A buyer who meets you in ChatGPT often arrives later by typing your name. Branded query growth in Search Console is real evidence the entity is landing.
• AI-referred sessions, monthly. Segment referrals from chatgpt.com, perplexity.ai, gemini.google.com, and copilot.microsoft.com in your analytics. The absolute numbers will be small compared with organic, but the conversion rate is usually not. Adobe Analytics measured generative AI traffic to US retail sites up 693% year over year in the 2025 holiday season, and traffic to tech and software sites up 120%, so the trend already reaches the categories our clients sell in.
• CRM attribution, quarterly. Add an AI assistant option to the how did you hear about us field and tag opportunities. This is the only number that ends the internal argument about whether the channel is real.

Expect the layers to move in order. Entity descriptions improve within weeks of the cleanup. Mentions on narrow prompts follow within one to three months, depending on how contested the category is. Branded search and AI-referred sessions lag by a month or two. Pipeline shows up last, and it shows up in the CRM as deals that name the assistant, which is how most of the results below were tracked.

What does entity SEO produce for B2B tech companies?

We run AI search optimization for software and tech companies and track outcomes against the CRM. Across the portfolio we hold an 80% success rate at getting a client recommended for a target commercial prompt, and the pattern is consistent: resolve the entity, own a narrow category, corroborate it, and the recommendations turn into pipeline.

The clearest example is Computools, a software development company. We positioned them as the recommended Salesforce partner inside the major LLMs, which is a textbook case of choosing one category and committing to it across every source. Within a three-month engagement they attributed $2M in deals to ChatGPT, including two enterprise contracts worth roughly $1M each.

They operated with the discipline and initiative of an internal senior marketer. (Computools, COO)

Baytech Consulting, also a software development company, reached a 100% placement rate across the AI-search prompts we targeted: recommended by every major AI assistant for three commercial keywords.

What impressed us most was their deep specialization in working with software development companies. (Baytech Consulting, Partner)

Speed depends on how narrow the category is. Opsworks, a DevOps company, was positioned for one important commercial keyword and recommended by the major AI assistants within a single month. SolarSpark, a small game development studio, was positioned in a niche AI-search category, recommended by LLMs for two commercial keywords within a month, and landed a large client in that same month.

They were not just talking about AI search in theory; they knew how to approach it practically. (SolarSpark, CEO)

The traffic and lead numbers follow the same logic. Gapsy Studio, a design agency, grew traffic from AI assistants from 10 to 154 sessions a month in three months, a 15x increase, after six months of nothing from a previous agency. Intelvision, a staff augmentation company, now generates two to four sales-qualified leads a month directly from ChatGPT alongside its paid program. Those are the AI-referred sessions and CRM attribution layers from the measurement stack above, showing up in real accounts. You can read the full set in our case studies.

What mistakes keep B2B tech companies from being recommended?

Most companies that are invisible in AI answers are making a handful of correctable entity mistakes. None are exotic, and because the field has not adapted yet, fixing them is often enough to pull ahead.

• Treating entity SEO as a schema task. Structured data declares who you are; it does not corroborate it. Organization markup on a site that nobody else describes consistently changes nothing.
• Letting the name and description fracture across platforms. Four versions of your company name and three versions of what you do give the engine four weak entities to choose from, so it chooses a competitor with one.
• Refusing to narrow. A company that claims six categories is associated strongly with none. The Semrush ownership data shows the open categories are the specific ones.
• Publishing adjectives instead of proof. World-class and results-driven are invisible to a model. Named clients, real numbers, and quotable client language are what get lifted and repeated.
• Ignoring the corroboration layer. No verified reviews, no roundup inclusion, no podcast or YouTube presence, and no community participation leaves the engine with nothing independent to repeat.
• Chasing citations and ignoring mentions. Being one of fifteen cited URLs is nice. Being one of three named vendors is the deal. Measure both, prioritize the mention.
• Measuring with SEO dashboards. Rankings cannot see the shortlist, so teams conclude the channel does nothing and defund the work that is producing pipeline.
• Doing it once. Models refresh, competitors invest, and stale facts erode the entity. Entity SEO is a cadence, not a launch.

Where does XQL fit?

Entity SEO and LLM citations are the core of our AI search optimization service for software and tech companies. We audit how the engines currently resolve your company, pick the category you can own, rebuild the corroboration layer across the sources each engine trusts, make your key pages quotable, ship the structured data, and run the prompt panel every week. Then we tie the whole thing back to pipeline in your CRM, because a recommendation that does not produce a deal is not a result we count.

If you are not sure how ChatGPT, Perplexity, or Gemini describe your company today, or whether you appear when your buyers ask for the best vendor in your category, we can check and show you the gap. Book a 30-minute intro call and we will map exactly where your entity stands and what it takes to get recommended.