Defenses

Prior-art research, old way vs new way — a side-by-side comparison

Defense is expensive because reading thousands of pages of patents takes time. AI doesn't change that for the lawyer's strategic work — but the time-on-pages part collapses by orders of magnitude.

6 min read · Updated Apr 29, 2026

Patent defense isn't expensive because lawyers are greedy. It's expensive because reading thousands of pages of patents and prior-art documents takes time — and time is the only thing senior associates actually sell. Most of the cost in a $300,000 invalidity workup is hours, not insight.

That's the cost AI changes. Not the strategy, not the courtroom, not the negotiation — those still need humans. But the legwork — search, summarize, cross-reference, claim chart — collapses from 80 hours to 3 minutes when you replace a senior associate with an LLM grounded in web search.

This is the side-by-side breakdown.

The Old Way: a senior associate, billing hourly

The traditional pre-litigation invalidity research engagement looks like this:

  • The defendant retains a white-shoe IP firm. Engagement letter, retainer.
  • A senior associate (typically a 4th-7th year, $400–$1,000/hour depending on firm and market) gets the file.
  • They pull the patent and read it: claims, specification, prosecution history. 8-15 hours.
  • They run searches in USPTO Patent Public Search and Google Patents for cited references and family members. 5-10 hours.
  • They expand the search to non-cited prior art using their judgment about scope. 20-40 hours.
  • They draft a claim chart mapping each limitation to candidate prior-art references. 15-30 hours.
  • They write a memo summarizing the strongest invalidity theories and any non-infringement angles. 10-20 hours.
  • A senior partner reviews the memo. 3-6 hours at $1,000-$1,500/hour.

Total billable hours: 60–120 for the associate, 3-6 for partner review.
Total cost: $35,000–$140,000 for a single patent.
Total elapsed time: 3-5 weeks.

The work is good. The associate is smart. The cost is real and unavoidable in the traditional model — the firm bills its inputs, not its outputs.

What this method doesn't do well:

  • Foreign-language prior art. Searching Japanese, Korean, or German databases requires either a multilingual associate or hiring a translator — both add time and cost.
  • Non-patent literature. Conference proceedings, vendor manuals, GitHub repos, archived web pages are all in scope under § 102, but searching them systematically adds 20+ hours an associate often doesn't have budget to spend.
  • Updates. The memo is point-in-time. If new prior art surfaces six months later, that's a new engagement.

The New Way: LLM with web-search grounding

The same six categories of analysis, run through a modern LLM (Claude Opus 4.7 or Gemini 2.5 Pro on this site) with grounded web search:

  • The system fetches the patent's full text from Google Patents.
  • The LLM is given the patent text as system context.
  • For each section — summary, litigation history, prior art (§ 102), obviousness (§ 103), term/extensions, derivative works — the LLM runs ~30-60 seconds of grounded web search across patent databases, academic literature, archived web pages, vendor documentation, and standards bodies. All languages.
  • It produces a structured markdown analysis with citations.
  • The analysis is published as a public web page indexed by Google and AI answer engines, queryable by anyone who later faces the same patent.

Total cost per patent: roughly $1–$5 in LLM API costs.
Total elapsed time: ~3 minutes for the full six-section dossier.
Updateability: re-run anytime; the search grounding picks up newer references with each refresh.

The work is good but not perfect. An LLM with web search is a junior associate that never gets tired, never has a budget cap, and reads every language. It does the legwork a senior associate would do — and surfaces the same kinds of patterns — but it doesn't make strategic calls.

Side-by-side

Dimension Old Way (white-shoe associate) New Way (LLM + web search)
Who does the work 4th–7th year associate, supervised by partner Claude Opus 4.7 or Gemini 2.5 Pro, grounded with live web search
Hourly cost basis $400–$1,000/hour for the associate; $1,000–$1,500 for partner review ~$0.30/minute in LLM API costs
Time per patent 3–5 weeks ~3 minutes for the full six-section dossier
Total cost per patent $35,000–$140,000 $1–$5
Sources searched USPTO; EPO/JPO if budget permits; English-language IEEE/ACM; whatever the associate has time to read Google Patents (170+ jurisdictions), Wayback Machine, GitHub, arXiv, IEEE/ACM, vendor manuals, standards bodies, conference proceedings, the indexed web
Languages covered English, with foreign-language searches as a paid extra All languages — the LLM translates and summarizes inline
Output structure One memo, ~30 pages, internal Word document Six independently-addressable sections: summary, litigation history, § 102 prior art, § 103 obviousness, extensions/term, defensive disclosure with derivative works + Mermaid diagrams
Citations Footnoted, internal Linked, public, dated, structured (Schema.org Article + LegalCase JSON-LD)
Cross-references Whatever cases the associate remembers; manual claim charts Auto-linked to the site's litigation database; every patent's dossier shows known cases, plaintiffs, defendants
Updateability Point-in-time. New prior art = new engagement Re-run anytime; minutes per re-run
Public availability Internal — only the client who paid sees it Public, indexed by Google and AI answer engines, citable by other defendants
Quality floor Depends on the associate's diligence and the budget Consistent — every patent gets the same systematic six-section treatment
Quality ceiling Capped by hours billed; senior partner review costs more Capped by what the LLM can find on the open web

Where the new way doesn't replace humans

This isn't an "AI replaces lawyers" article. The new way replaces one specific kind of work: open-ended factual research. Everything else still needs human judgment:

  • Claim construction is a judgment call grounded in case law and the prosecution history. AI helps you map; the lawyer argues.
  • Strategy — file IPR or stay in district court? Pursue § 101 first or § 103? Settle or fight? — needs experience, not pattern matching.
  • Trial work — depositions, witness prep, courtroom advocacy — is irreducibly human.
  • Settlement negotiation depends on relationships, leverage reading, and judgment about the other side.
  • Filing legal papers — IPR petitions, motions, complaints — requires a licensed attorney. Period.

The right framing: AI does the work that used to consume 80 of the 100 hours in a typical engagement. A lawyer still does the strategic 20 hours — but the engagement that used to cost $80,000 now costs $20,000, because the cheap-and-tedious 80 hours is delivered for the price of a few API calls.

For defendants who can afford a $20,000 engagement but not an $80,000 one, that gap is the difference between fighting and folding. (How patent trolls operate covers why folding is exactly what the troll wants.)

The economics this enables

The reason patent trolls profit from junk patents is that defending costs more than settling. (See the asymmetry math.)

Once invalidity research collapses from $80k to $5, the whole asymmetry shifts. A defendant who would have settled for $200k now considers spending $20k on a real defense. Some fraction of those defendants win clean, get fees back under § 285, and the troll's expected-value math turns negative on the next campaign.

The pre-built dossier is the structural fix. Once a patent has a public invalidity record, the troll's per-target settlement value drops because every defendant starts the engagement at $0 of prior-art research instead of $50,000. The math that lets the troll campaign profit from junk patents only works when defendants are uncoordinated and starting from scratch. AI-generated dossiers fix the second problem — and once enough exist, the first follows.

Bottom line

The old way wasn't wrong; it just priced defense out of reach for everyone but Fortune-500 companies. The new way doesn't make lawyers obsolete — it makes the cheap-and-tedious 80% of their old work obsolete, so they can focus on the strategic 20%. Defendants get the same coverage at a fraction of the cost. Trolls lose the per-defendant prior-art tax that funded the model.

If you're a defendant: browse the database for the patent that's been asserted against you. If we don't have it, add it — verification + a six-section analysis run in about 3 minutes. If your lawyer is starting prior-art research at $600/hour, hand them the dossier and ask them to start at the strategy work instead.

This article is for general education and is not legal advice. AI-generated invalidity analyses are a starting point, not a substitute for the judgment of a licensed patent attorney.