Original research · 2026-07 edition

AI SEO Statistics: Intellectual Property (2026-07 edition)

40 questions · 120 AI responses · 3 models · measured 2026-07-06

The question bank

The questions we tested — sampled from real buyer journeys in intellectual property.

Each model answered every question once, same wording, same day. These are the prompts behind every percentage on this page.

How do I know if my invention is actually patentable before I spend money on a lawyer?
What's the difference between a trademark and a service mark for my new consulting business?
Can I handle a provisional patent application myself or will I mess up the filing date?
I just received a cease and desist letter for my Etsy shop, what should my first step be?
How much should I expect to pay for a full trademark search and filing in the US?
Is it better to hire a boutique IP firm or a large full-service law firm for a tech startup?
What are the red flags to look for when interviewing a patent attorney?
Do I need a patent agent or a patent attorney if I just need help with the technical drawings?
Show all 40 questions
How do I protect my brand name internationally if I'm planning to sell on Amazon Europe?
What's the typical timeline from filing a trademark application to getting it approved?
Should I file for a utility patent or a design patent for my new kitchen gadget?
How do I prove someone infringed on my copyright if they changed a few things in the design?
Is a trade secret better than a patent for a proprietary software algorithm?
What questions should I ask during a free consultation with an IP lawyer?
How do I find an IP lawyer who specializes in the biotech industry?
Is it worth hiring a lawyer to respond to a USPTO office action or can I do it myself?
How much does it cost to defend a patent in court if a competitor copies my product?
What happens if I launch my product before my patent application is officially filed?
Can I lose my trademark if I don't use it for a certain period of time?
Do I need a local IP lawyer or can I work with someone remotely across the country?
How do I transfer ownership of a patent I bought from another inventor?
What's the average hourly rate for an intellectual property lawyer in a major city?
Someone is using my logo on social media, can an IP lawyer get their account banned?
How do I protect my intellectual property when I'm hiring freelance developers?
Is there a way to speed up the patent examination process with the USPTO?
What's the risk of using a cheap online filing service instead of a specialized IP attorney?
How do I conduct a freedom to operate search before launching a new manufacturing line?
Can I trademark a phrase that is commonly used in my industry?
What should be included in a standard non-disclosure agreement to protect my invention?
How do I handle a situation where a former employee stole my customer list and trade secrets?
Is it possible to get a patent for a business method or just for physical inventions?
What are the common reasons why a trademark application gets rejected by the examiner?
How do I budget for patent maintenance fees over the next ten years?
Should I hire an IP lawyer to draft my licensing agreement or use a template?
What's the difference between a patent search and a patentability opinion?
How do I verify if a patent attorney is actually registered with the USPTO?
Can I sue for copyright infringement if I haven't officially registered my work yet?
How do I protect my brand if someone buys a domain name that is my trademarked business name?
What are the pros and cons of filing a PCT application for international patent rights?
How do I know if my IP lawyer has a conflict of interest with my competitors?

Model by model

18-point average divergence: which AI you ask changes the answer.

The divergence index is the average gap between the most and least likely model per behavior. Higher = the models disagree more about intellectual property buyers.

Behavior rates across 40 intellectual property buyer questions, 2026-07 edition. Last column: average across models.
ChatGPTClaudeGeminiConsensus
Recommends hiring a professional95%78%28%33%
Suggests DIY first50%33%18%63%
Names specific providers5%10%3%83%
Gives price or cost info18%23%13%73%
Tells to check reviews8%3%3%90%
Tells to verify credentials13%5%8%88%
Mentions case studies / portfolio13%0%0%88%
Mentions local proximity5%8%8%90%
Gives selection criteria15%23%20%65%
Warns about red flags15%13%8%85%
Asks a clarifying question53%65%0%25%
Recommends multiple quotes3%3%0%95%

By model

How each assistant handled Intellectual Property questions.

Reading the 120 answers model by model shows how differently the three assistants treat the same intellectual property questions. On the most consequential behavior — whether to send the buyer to a professional at all — the rate ranged from 95% (ChatGPT) down to 27.5% (Gemini), a 68-point gap on an identical question set.

Across the 40 intellectual property answers it produced, ChatGPT recommended hiring a professional in 95% of them and suggested a DIY approach first 50% of the time. It named a specific provider in 5% of answers (about 0.1 distinct providers per answer) and included price or cost information 17.5% of the time. ChatGPT asked a clarifying question before answering in 52.5% of cases, warned about red flags or scams in 15%, and told the buyer to verify credentials in 12.5%, averaging 624 words per answer. On the remaining cues it told the buyer to check reviews in 7.5%, pointed to case studies or a portfolio in 12.5%, and framed the choice around local proximity in 5%; a selection-criteria checklist appeared in 15% of its answers and a recommendation to gather multiple quotes in 2.5%.

Across the 40 intellectual property answers it produced, Claude recommended hiring a professional in 77.5% of them and suggested a DIY approach first 32.5% of the time. It named a specific provider in 10% of answers (about 0.3 distinct providers per answer) and included price or cost information 22.5% of the time. Claude asked a clarifying question before answering in 65% of cases, warned about red flags or scams in 12.5%, and told the buyer to verify credentials in 5%, averaging 331 words per answer. On the remaining cues it told the buyer to check reviews in 2.5%, pointed to case studies or a portfolio in 0%, and framed the choice around local proximity in 7.5%; a selection-criteria checklist appeared in 22.5% of its answers and a recommendation to gather multiple quotes in 2.5%.

Across the 40 intellectual property answers it produced, Gemini recommended hiring a professional in 27.5% of them and suggested a DIY approach first 17.5% of the time. It named a specific provider in 2.5% of answers (about 0.1 distinct providers per answer) and included price or cost information 12.5% of the time. Gemini asked a clarifying question before answering in 0% of cases, warned about red flags or scams in 7.5%, and told the buyer to verify credentials in 7.5%, averaging 249 words per answer. On the remaining cues it told the buyer to check reviews in 2.5%, pointed to case studies or a portfolio in 0%, and framed the choice around local proximity in 7.5%; a selection-criteria checklist appeared in 20% of its answers and a recommendation to gather multiple quotes in 0%.

Taken together, ChatGPT is the assistant most likely to route an intellectual property buyer to a professional (95%) and Gemini the least (27.5%). ChatGPT produced the longest answers, at 624 words on average. Specific providers were named most often by Claude (10%) — even there, roughly one answer in 10 carried a name.

Where they disagree

The behaviors where the choice of model changes the answer.

The divergence index for this study is 18.1 points — the average distance between the most and least likely model across the coded behaviors. The gaps below are where which assistant an intellectual property buyer happens to ask matters most:

  • Recommends hiring a professional: from 27.5% (Gemini) to 95% (ChatGPT) — a 68-point spread.
  • Asks a clarifying question: from 0% (Gemini) to 65% (Claude) — a 65-point spread.
  • Suggests a DIY approach first: from 17.5% (Gemini) to 50% (ChatGPT) — a 33-point spread.
  • Mentions case studies or portfolio: from 0% (Claude) to 12.5% (ChatGPT) — a 13-point spread.
  • Gives price or cost information: from 12.5% (Gemini) to 22.5% (Claude) — a 10-point spread.

The widest single gap — recommends hiring a professional, 68 points — means an intellectual property buyer can receive materially different guidance on the same question depending only on which assistant they happen to open, so any visibility strategy built on a single model's behavior describes only part of the intellectual property market.

Where they agree

The points of near-consensus in Intellectual Property.

On other behaviors the three models move almost in lockstep — the points of near-consensus for intellectual property, where all three landed within a few points of each other:

  • Mentions local proximity: 5%–7.5% across all three (a 3-point spread).
  • Recommends multiple quotes: 0%–2.5% across all three (a 3-point spread).
  • Tells the buyer to check reviews: 2.5%–7.5% across all three (a 5-point spread).
  • Names a specific provider: 2.5%–10% across all three (a 8-point spread).

Measured question by question, the three assistants coded a response the same way most consistently on "recommends multiple quotes" (identical coding in 95% of questions) and least consistently on "asks a clarifying question" (25%).

Every behavior, measured

All twelve coded behaviors for Intellectual Property, averaged across the three models.

The behaviors AI models reproduce most often for intellectual property are recommends hiring a professional (66.7% on average), asks a clarifying question (39.2%) and suggests a DIY approach first (33.3%); the rarest are recommends multiple quotes (1.7%), mentions case studies or portfolio (4.2%) and tells the buyer to check reviews (4.2%). Each figure below is the share of a model's 40 answers in which the behavior appeared at least once, averaged across the 3 models with the full per-model range in parentheses:

  • Recommends hiring a professional: 66.7% on average (ChatGPT 95%, Claude 77.5%, Gemini 27.5%) — a 68-point spread.
  • Asks a clarifying question: 39.2% on average (ChatGPT 52.5%, Claude 65%, Gemini 0%) — a 65-point spread.
  • Suggests a DIY approach first: 33.3% on average (ChatGPT 50%, Claude 32.5%, Gemini 17.5%) — a 33-point spread.
  • Gives selection criteria: 19.2% on average (ChatGPT 15%, Claude 22.5%, Gemini 20%) — a 8-point spread.
  • Gives price or cost information: 17.5% on average (ChatGPT 17.5%, Claude 22.5%, Gemini 12.5%) — a 10-point spread.
  • Warns about red flags or scams: 11.7% on average (ChatGPT 15%, Claude 12.5%, Gemini 7.5%) — a 8-point spread.
  • Tells the buyer to verify credentials: 8.3% on average (ChatGPT 12.5%, Claude 5%, Gemini 7.5%) — a 8-point spread.
  • Mentions local proximity: 6.7% on average (ChatGPT 5%, Claude 7.5%, Gemini 7.5%) — a 3-point spread.
  • Names a specific provider: 5.8% on average (ChatGPT 5%, Claude 10%, Gemini 2.5%) — a 8-point spread.
  • Tells the buyer to check reviews: 4.2% on average (ChatGPT 7.5%, Claude 2.5%, Gemini 2.5%) — a 5-point spread.
  • Mentions case studies or portfolio: 4.2% on average (ChatGPT 12.5%, Claude 0%, Gemini 0%) — a 13-point spread.
  • Recommends multiple quotes: 1.7% on average (ChatGPT 2.5%, Claude 2.5%, Gemini 0%) — a 3-point spread.

Trust signals

How well the models protect the intellectual property buyer.

Beyond whether to hire, the rubric codes how carefully each assistant protects the intellectual property buyer once a decision is made. Telling the buyer to check reviews or ratings appeared in 4.2% of answers on average. Verifying credentials or certifications appeared in 8.3%. Warning about red flags or scams appeared in 11.7%.

On structuring the decision, a selection-criteria checklist showed up in 19.2% of answers on average and a recommendation to gather multiple quotes in 1.7%. The single least-reproduced protective signal for intellectual property is "recommends multiple quotes" at 1.7% on average — the clearest opening for content that supplies it, since the models are not yet reliably surfacing that guidance on their own.

Referral behavior

Do AI models name Intellectual Property providers?

For service providers the decisive question is whether these systems name anyone at all. Across 120 intellectual property answers, a specific provider was named in 5.8% of responses on average — roughly 0.2 distinct providers per answer. In practice the assistants behave far more as an explanatory layer than as a referral engine for intellectual property: visibility comes from being the reasoning a model reproduces, not from being the named recommendation.

The question set

What these 40 Intellectual Property questions cover.

The 40 questions behind every percentage on this page were drawn from real intellectual property (legal services; buyer hiring decisions for this specific service) buyer journeys. Each was put to all 3 models once, with identical wording, so the rates above describe how the assistants handled this exact intellectual property question set — not a general prior or a hand-picked subset. The full list is shown earlier on this page; the coded percentages are what those specific questions produced.

How to read this

A note on the numbers.

A percentage here is the share of a model's 40 answers in which the behavior appeared at least once — not a confidence score. Because each model answered every question exactly once on 2026-07-06, the figures describe this specific intellectual property question set and snapshot rather than a general prior. The full protocol and coding rubric are documented in the study methodology.

Methodology

A controlled snapshot, documented end to end.

40 standardized buyer questions per industry, one response per model per question (ChatGPT (gpt-5-mini), Claude (claude-sonnet-5), Gemini (gemini-3-flash-preview)), collected 2026-07-06, coded against a fixed 12-behavior rubric with human QA. AI outputs vary with model version, location and time — figures describe this sample and window, and are refreshed each edition. Read the full methodology →