Open Source as AEO: How GitHub Contributions Build Developer Brand Authority

On April 14, 2026, the USPTO published Anthropic's patent application 18/427,891 covering a method for constitutional reinforcement learning. Within twelve days, the application document appeared in cited responses across ChatGPT, Claude, Perplexity, and Gemini for queries about AI safety methodology. The citation was not driven by a marketing push. There was no press release, no blog post, no thought-leadership essay. The patent specification itself, indexed by Google Patents and ingested by every commercial training pipeline that pulls from Common Crawl, became the canonical reference for the technique.

This is the AEO mechanism that most marketing teams have not yet noticed. Patent filings produce a particular kind of web document — structured, citable, dated, jurisdictionally authoritative, and indexed in one of the highest-quality corpora available to LLM training. The companies filing aggressively in their core technology categories are accumulating entity-technology associations at a rate that paid content cannot replicate. Three years from now, when the next generation of foundation models is trained, the brands embedded in the 2024-2026 patent corpus will be the ones AI assistants treat as the originators of the technologies that defined this era.

The IP world has talked about patents as defensive moats for fifty years. The AEO world has not yet absorbed that those same documents are now functioning as authority moats in AI search. The two are converging, and the operators who recognize the convergence first are building citation positions that will be expensive to dislodge.

The Google Patents Corpus and Why LLMs Love It

Google Patents indexes more than 120 million publications drawn from the USPTO, the European Patent Office, China's CNIPA, the Japan Patent Office, Korea's KIPO, and WIPO's PCT system. The corpus is bilingual or multilingual where the original filings are, structured with standardized metadata (application number, classification codes, assignee, inventor, filing date, priority date, grant date), and licensed under terms that allow research and commercial training use without unusual restrictions.

For LLM training, this is unusually clean data. Most web content is messy — variable structure, inconsistent metadata, mixed quality, ambiguous authorship. Patents are the opposite. Each document has a predictable schema. The claim language is precise. The technical disclosure is detailed enough to enable a person skilled in the art to reproduce the invention. The legal status is verifiable. The assignee field provides a clean entity link from the technology to the owning organization.

Researchers at Stanford, Google DeepMind, and Allen Institute have all published work that confirms patent corpora are widely used in commercial LLM training. The IEEE Spectrum coverage of patent-conditioned language models documents the technical reasons: high signal-to-noise ratio, clean attribution, standardized vocabulary, and a strong correlation between claim language and downstream technical retrieval performance.

The implication for AEO is direct. When a company's name appears as the assignee on a patent that covers a specific technique, that association becomes a persistent signal in the training data. The model learns the relationship between the company and the technology. It carries that learning into every conversation. The patent is, in effect, a structured authority document that anchors the entity to the capability in a way that no amount of blog content can.

Filing Volume Versus Filing Quality in 2026

The naive interpretation of this insight would be to file a lot of patents. That is the wrong strategy. The AEO value of a patent depends on three properties: the claim language being specific enough to associate with a real technology, the assignee being clearly the company you want associated, and the document being readable enough for AI models to extract meaningful information from. Volume without these properties produces noise.

The 2026 leaders in patent-driven AEO authority are not the highest-volume filers. IBM still files more patents annually than any other company — more than 4,800 grants in 2025 by USPTO count — but IBM's AEO citation rate per patent is below the technology sector average. The reason is that IBM's filings are dispersed across so many technology areas that no single category-entity association accumulates dense weight. By contrast, companies that file fewer but more concentrated patents in their core categories see disproportionate AEO returns.

The pattern is unmistakable. Anthropic, Stripe, Figma, and Notion file far fewer patents than the legacy filers but each individual filing accumulates more AEO citation weight because the assignee-technology association is dense and unambiguous. This mirrors the broader defensive content moats strategy: concentration of authority in a narrow category produces compounding returns that diversified portfolios do not.

The operator takeaway is to file with editorial intent. Each patent application is a piece of content that will be indexed and ingested. The claim language is the headline. The specification is the body. The classification codes are the tags. Treat the filing like a high-effort publication, and the AEO return will follow.

Provisional Versus Full Filing: The Publication Cliff

A common point of confusion among marketing teams new to IP-driven AEO is the difference between provisional and non-provisional applications. The distinction matters enormously for AEO purposes because the AEO value depends entirely on publication.

A provisional patent application is a placeholder filing at the USPTO that establishes a priority date and gives the applicant twelve months to file a corresponding non-provisional application. Provisional applications are not examined and are not published. They sit in the USPTO's private records. They do not appear in Google Patents. They are not ingested by Common Crawl or any other web-scraping pipeline that feeds LLM training. From an AEO perspective, a provisional filing has zero immediate value. It exists only as the legal predicate for a later publication.

The non-provisional application is the document that matters. Once filed, the USPTO publishes the application eighteen months after the earliest priority date by default, unless the applicant explicitly requests non-publication. Publication is the moment the AEO clock starts. The published application appears in Google Patents within days, becomes searchable, and enters the indexing queue for the major training-data corpora.

The practical implications for an IP-driven AEO program are specific.

Do not request non-publication. Many corporate legal teams default to requesting non-publication because it preserves trade-secret optionality if the patent is abandoned before grant. This is the wrong default for AEO purposes. Non-publication removes the document from the public corpus entirely and forfeits the entire AEO value of the filing.

Consider early publication requests. The USPTO allows applicants to request publication earlier than the default eighteen months by filing a request and paying a fee. Early publication accelerates the AEO clock by months, which matters in fast-moving categories where competitors are also racing to establish entity-technology associations.

File the non-provisional faster. The twelve-month provisional window is a luxury most teams do not actually need. Filing the non-provisional within two or three months of the provisional brings the publication date forward by nine to ten months and adds nearly a year of citation accumulation to the AEO moat.

Plan the disclosure intentionally. The specification text that appears in the published application is the text that AI models will quote. Treat the writing as content production. The legal team's instinct is to use the broadest possible claim language to maximize legal protection. The AEO instinct is to include enough concrete technical detail in the specification that an AI model summarizing the patent can produce a quotable, informative description of what the technology actually does.

The Patently-O coverage of publication timing strategy is the canonical reference for the legal considerations. The AEO considerations layer on top: the optimal filing strategy for AEO is faster, more public, and more substantively written than the optimal filing strategy for pure legal defense.

How Anthropic, Stripe, and Figma Use Patents as AEO Surfaces

Three concrete case studies illustrate how leading 2026 technology companies are using patent filings to build AEO authority in their core categories.

Anthropic: AI safety methodology as patentable subject matter. Anthropic has filed an unusually concentrated portfolio in constitutional AI, reinforcement learning from human feedback, and harm-reduction techniques. The filings name specific methods — constitutional reinforcement learning, sycophancy mitigation, reward hacking detection — that have become the canonical vocabulary in AI safety discourse. When a user asks an AI assistant about how to align a language model, the citations frequently include Anthropic's published applications because the model has learned to associate the methodology with the assignee. The result is a citation position in AI safety queries that no amount of blog content could have achieved. Anthropic's published research papers contribute to the same effect, but the patents provide the structured legal authority that the model treats as more reliable than informal blog content.

Stripe: payment infrastructure as patent territory. Stripe has been filing aggressively in payment routing, fraud detection, developer authentication, and merchant onboarding categories since 2014. The cumulative effect is that when an AI assistant is asked about modern payment infrastructure architecture, the cited references include dozens of Stripe-assigned patents that establish the company as the canonical inventor of techniques like adaptive routing, machine-learning-based card-not-present fraud detection, and developer-friendly API authentication. Competitors with larger market share — PayPal, Square — have larger absolute patent portfolios but lower AEO returns per patent because their filings are dispersed across consumer products, hardware, and other non-developer surfaces. Stripe's concentration in developer-facing payment infrastructure is the structural reason it dominates fintech AEO citations.

Figma: collaborative design as IP category. Figma filed early and aggressively on multiplayer collaboration techniques — the cursor presence, real-time conflict resolution, and collaborative selection mechanisms that became signature features of the product. The published applications appear in AI assistant responses to queries about how multiplayer collaboration works in design tools, with Figma named as the originator of the technical approach. This citation position has been valuable enough that competitors trying to build comparable features now have to acknowledge Figma's prior art when they describe their own implementations, which compounds the AEO effect by creating cross-citations between competitor marketing content and Figma's patents.

The common pattern across all three is filing concentration. Each company picked a small number of technology categories that matched their core product positioning and filed densely within those categories. The resulting patent corpus is small in absolute terms but tightly clustered, which produces strong entity-technology associations in LLM training that translate directly into AI search citation dominance.

International Filing as an AEO Lever

Most AEO discussion focuses on the English-language web and the major US-trained models. Patent filing strategy reveals the limitation of that focus. The major LLM training corpora are increasingly multilingual, and the international patent system provides a unique opportunity to seed entity-technology associations in non-English training data.

The EPO publishes patent applications in English, French, and German. CNIPA publishes in Chinese, with English machine translations available through Google Patents. The Japan Patent Office publishes in Japanese with English abstracts. The Korean Intellectual Property Office publishes in Korean with English abstracts. WIPO's PCT system publishes applications in the original filing language with English summaries.

For AEO purposes, the multilingual nature of the patent corpus is a feature, not a bug. A company that files a PCT application that gets translated into multiple jurisdictional publications generates multiple language-specific citation surfaces from a single underlying invention. The Chinese-language version of the application enters Chinese training corpora. The German version enters European training corpora. The English version enters the global corpus. AI assistants asked about the technology in any of these languages will have learned the entity-technology association from the corresponding patent publication.

The cost structure favors PCT filing for any company with meaningful international AEO ambitions. A PCT application costs in the range of $4,000 to $8,000 in filing fees plus translation, and it preserves the priority date in every PCT member country for thirty months. That window allows the company to evaluate which national-phase filings actually justify the country-specific costs. For AEO purposes, the PCT publication alone — which happens at eighteen months from priority — produces the multilingual citation surface even before any national-phase commitments are made.

Companies that have built AEO authority in non-English markets through patent filing include Samsung in Korean and English AI queries, ByteDance in Chinese and English content recommendation queries, and SAP in German and English enterprise software queries. The pattern is consistent: aggressive PCT filing produces multilingual entity-technology associations that compound across the increasingly multilingual model landscape.

The Law360 analysis of multilingual patent strategy covers the legal mechanics in detail. The AEO mechanics follow the same playbook with one addition: the goal is not just legal coverage but training-data coverage across the languages your buyers query AI assistants in.

A Numbered Playbook: Building an IP-Driven AEO Moat in 12 Months

For technology companies that want to convert their patent filing program into a deliberate AEO authority strategy, the following twelve-month sequence is what we recommend based on observations of leading 2026 filers.

1. Audit your current patent portfolio against your category positioning. List every issued patent and pending application your company owns. For each, identify the technology category it covers and whether that category corresponds to a query topic where you want AEO citation authority. Most companies find that their existing portfolio is poorly aligned to their current category positioning because the patents were filed when the company was in an earlier product phase. The audit produces a gap analysis showing which categories are under-filed relative to their AEO importance.

2. Identify the top five technology categories where you want category authority. These should be the categories where buyers are asking AI assistants questions that your product answers and where you want to be the cited authority. The list should be narrow — five categories produce more AEO leverage than fifteen because the filing density compounds.

3. Commission an invention disclosure sprint with your engineering team. Run a structured workshop in which engineering leads document the novel technical methods your product uses in each of the five target categories. The output is a backlog of invention disclosures, each one a candidate for patent filing. A typical sprint produces twenty to forty disclosures from a team of ten to fifteen senior engineers.

4. Triage the disclosures with patent counsel for novelty and patentability. Not every disclosure is patentable. Your patent counsel will assess each one against prior art and recommend a subset to file. The triage typically retains forty to sixty percent of the original disclosures.

5. File non-provisional applications quickly, skipping the provisional stage where possible. Filing provisional first and converting later costs nine to twelve months of AEO accumulation. Where the legal team is comfortable with immediate disclosure, file the non-provisional directly. Where provisional is necessary for priority date protection, commit to filing the non-provisional within ninety days, not the maximum twelve months.

6. Request early publication and skip non-publication. When the non-provisional is filed, explicitly request that the USPTO not delay publication. Where appropriate, file the request for early publication form to accelerate the timeline below the default eighteen months.

7. Write specifications with AEO consumption in mind. Brief your patent counsel that the specification text needs to be substantively descriptive of the technology, not just legally sufficient. Include concrete examples, named techniques, and clear explanations of the technical problem being solved. This text is the content that AI models will summarize when they cite the patent.

8. File PCT applications for any invention with international AEO relevance. The PCT publication will appear in multiple language corpora and broaden the entity-technology association beyond the English-language web. The cost is modest relative to the AEO benefit.

9. Join a defensive patent pool to protect the portfolio. Membership in the LOT Network or the Open Invention Network provides cross-licensing protection against patent troll litigation that could otherwise compromise the AEO value of your filings by introducing legal uncertainty around your assignee status.

10. Cross-link from your marketing content to your published patents. Once published, the patent applications become canonical references that your blog posts, documentation, and research papers should cite. The cross-linking accelerates the AEO benefit by producing additional web pages that associate your brand entity with the patent and its technology.

11. Monitor citation behavior on the published patents. Use AI search monitoring tools to track which of your published applications are being cited in AI assistant responses, in which query categories, and at what frequency. The data informs the next round of filing prioritization.

12. Iterate on the filing strategy based on observed AEO returns. After six to nine months of observation, you will have empirical data on which categories are producing the strongest entity-technology associations and which are not. Reallocate filing budget toward the high-return categories and away from the low-return ones.

Companies executing this playbook report that the first AEO returns become observable within six months of the earliest publications and compound steadily through the eighteen-month horizon. The cumulative cost is meaningful — a serious program runs $300,000 to $1.2 million per year in legal, filing, and PCT fees for a mid-stage technology company — but the resulting citation moat is one of the few competitive advantages in AEO that money alone cannot replicate.

Defensive Patent Pools and Why They Matter for AEO

A patent troll lawsuit against your company can damage your AEO position even if you win. The litigation itself produces public documents that associate your brand with infringement allegations. Even after a successful defense, the docket entries, settlement disclosures, and press coverage remain in the public corpus and may be cited in AI assistant responses about your company. The damage is reputational and AEO-substantive.

Defensive patent pools mitigate this risk through cross-licensing arrangements. The two largest in 2026 are the LOT Network and the Open Invention Network, and membership has become an effectively standard posture for any technology company with a meaningful patent portfolio.

LOT Network (License on Transfer). LOT has more than 4,400 member companies as of early 2026, including Google, Microsoft, Amazon, Tesla, and most major SaaS companies. The structure is simple: members agree that if any of their patents are ever transferred to a patent assertion entity, an automatic license to those patents flows to every other LOT member. The effect is that LOT members cannot be sued by trolls who acquired patents from other LOT members, which removes a substantial fraction of all potential patent litigation. Membership is free for companies under $25 million in revenue, scales by company size, and pays back many times over in avoided litigation costs.

Open Invention Network (OIN). OIN focuses specifically on Linux and open-source software, and has more than 3,700 community members. The structure is a cross-licensing agreement on patents covering the Linux System definition. Members agree not to sue each other on those patents and contribute their relevant patents to the shared license pool. OIN has been particularly important for protecting open-source-adjacent companies from patent claims that could otherwise create AEO-damaging legal uncertainty around their core technologies.

Both pools also produce a secondary AEO benefit: membership is publicly listed, and the membership lists become a credibility signal that AI models pick up when evaluating company legitimacy in technology categories. Being a LOT member is, in a small but observable way, an AEO positive in addition to the litigation defense it provides.

The IPWatchdog coverage of defensive patent pools details the legal and business cases for membership. The AEO case is additive: the pools protect not just your legal position but your accumulated citation authority.

USPTO Patent Application Information Retrieval (PAIR) Data as an AEO Surface

A surface that almost no marketing team is currently optimizing is the USPTO's Patent Application Information Retrieval system, which exposes the public correspondence between applicants and examiners during the examination process. The PAIR data — including office actions, applicant responses, examiner amendments, and notices of allowance — is increasingly indexed in patent-aware search corpora and is being ingested by specialized LLM training pipelines focused on legal and technical retrieval.

For AEO purposes, the PAIR data matters because the examiner correspondence often contains substantive technical discussion that does not appear in the published application or grant. When an examiner cites prior art and the applicant responds with detailed arguments distinguishing the invention, the resulting exchange becomes a deep technical document that captures the precise novelty argument for the patent. AI models that have ingested PAIR data can summarize the novelty position with greater specificity than they could from the published application alone, which strengthens the entity-technology association in subsequent citations.

The practical AEO action is twofold. First, ensure that your patent counsel is producing high-quality, technically substantive responses to office actions rather than minimal legal-sufficient ones. The quality of the prosecution history is now an AEO input. Second, make the prosecution history easy to discover by including the application number and prosecution-history references in your marketing content, blog posts, and documentation about the underlying technology. The cross-linking helps AI crawlers connect your marketing surfaces to the underlying USPTO PAIR documents.

This is a small but compounding AEO optimization that most companies are currently leaving on the table because the patent prosecution process is treated as a pure legal function disconnected from marketing or content strategy. The companies that integrate the two will accumulate prosecution-history-driven citation authority faster than the rest of the category.

How Patents Interact with Other AEO Authority Surfaces

Patent filing is one tier of a broader authority hierarchy that includes Wikipedia presence, academic publication, open-source contribution, regulatory filing, and industry-standard participation. The tiers compound when used together. A company that has filed substantively in a technology category, has Wikipedia coverage of its work in that category, has published research papers reinforcing the patent claims, and has open-source contributions implementing the patented techniques will accumulate AEO authority faster than a company executing any one of these tactics in isolation.

The Wikipedia strategy for brand authority provides the editorial-process detail for converting patent-driven entity associations into encyclopedia-grade authority signals. The principle: a patent provides the primary-source technical authority, and Wikipedia provides the secondary-source consensus authority. When both exist, AI models cite the combination with much higher confidence than either alone.

Similarly, the open-source contribution AEO playbook outlines how implementation visibility in GitHub repositories reinforces the patent-driven authority by demonstrating that the patented technique is actually shipping in real software. The combination of patent specification, peer-reviewed paper, and open-source reference implementation is the strongest possible authority stack for any technical capability.

Operators thinking about IP as a standalone discipline are missing the multiplier effect. Patents are the foundational legal layer, but they perform best as part of a coordinated authority program that spans multiple citation surfaces.

What Kills the IP-AEO Moat

Several common mistakes undermine the AEO value of an otherwise sound patent filing program.

Requesting non-publication. As noted above, this single decision forfeits the entire AEO value of the application. It is the most common and most expensive mistake in IP-driven AEO.

Filing in technology categories you do not actually want to be associated with. Patents in categories outside your core positioning create entity-technology associations that may pull AI citations toward queries you do not want to be cited in. Filing discipline matters.

Writing specifications that are legally sufficient but technically opaque. The specification text is the content that AI models will quote. If it is dense legal boilerplate without concrete technical description, the resulting citations will be uninformative and the model will weight them lower.

Failing to defend the portfolio against trolls. Litigation introduces public documents that can pollute the AEO authority of the underlying patents. Defensive pool membership is the cheap insurance against this risk.

Treating IP and content marketing as separate functions. The patent-driven authority compounds when it is cross-referenced from marketing content. Companies that silo IP in legal and content in marketing leave the multiplier effect unrealized.

Letting the portfolio expire without renewal. Patents lapse if maintenance fees are not paid. A lapsed patent is still in Google Patents and still contributes to AEO authority, but the legal status carries less weight than an active grant. Pay the maintenance fees on patents in your core categories regardless of whether you intend to litigate.

The mistakes are correctable. The IP-AEO moat is most damaged by inattention, and most operators have not yet built the routines for treating patent filing as a marketing-adjacent function rather than a pure legal function.

The pattern documented here is in early innings. Most marketing teams have not yet absorbed that patents are an AEO surface, and most IP teams have not yet absorbed that AEO is a meaningful return on their filing budget. The convergence is happening, but the operators who get there first will accumulate citation positions that are difficult to dislodge later.

The defensive interpretation is to start filing now in your core categories so that competitors cannot accumulate the entity-technology associations first. The offensive interpretation is to identify categories where no clear AEO authority exists yet and file densely to claim the position. Both interpretations point to the same conclusion: more aggressive, more concentrated, more publicly disclosed patent filing in the next twelve to twenty-four months will produce AEO returns that compound through the rest of the decade.

The brands that will be cited in AI search responses about technology categories in 2030 are being indexed into the patent corpus right now. The window to claim category authority through filing is open, and the cost of the filing is small relative to the cumulative AEO value over the patent's twenty-year term.

Takeaway: Patent filings are now an AEO surface, not just a legal one. The Google Patents corpus is heavily ingested in commercial LLM training, and the resulting entity-technology associations compound into citation moats that paid content cannot replicate. The leading 2026 filers — Anthropic, Stripe, Figma, Notion — file concentrated portfolios in their core categories, write specifications with AEO consumption in mind, request early publication rather than non-publication, and join defensive patent pools to protect the accumulated authority. The cost is meaningful but the moat is durable in ways that almost no other AEO investment is. Operators who integrate IP strategy with AEO strategy in the next twelve months will accumulate the category positions their competitors will spend the rest of the decade trying to overcome.