11 Rules for US vs UK Patentability of AI-Driven Personalized Nutrition (2025)

Not legal or medical advice. This article is for education and planning. Talk to qualified counsel and clinicians before acting.

Expert review: External Patent Counsel (name on file), Registered Patent Attorney. Review date: 2025-10-09 (KST).

File Today, Publish Tomorrow: A Clean Sequence That Survives US/UK Review

You’re being pulled in three directions at once. We can move now, keep publication open, and avoid boxing in the roadmap—like keeping the kettle on while you set the table.

However, in the US, claims that read like “recommendations” risk a 35 U.S.C. §101 abstract-idea rejection; in the UK, outputs framed as non-technical “taste” or preference struggle under the technical-contribution test. So we anchor everything to measurable steps with clear cause-and-effect.

I once watched a founder sign the POA at 16:50 in a noisy café, the espresso machine hissing beside us; what saved us wasn’t speed, it was order.

• Ground the invention in a technical pipeline. Describe it with 4–6 verbs: sample → filter → normalize → extract → threshold → output. Replace “recommend” with concrete actions: detect, transform, classify, actuate, display state. Example: “If noise < 5% and feature X ≥ 2.0, display state Y for 30 s.” (So an examiner can verify behavior, quietly and directly.)
• Separate filing from storytelling. File a provisional naming human inventors and include 2–3 working examples with small, real numbers. Let the “metabolic reset” page explain user flow, not outcomes or promises; we won’t publish a preprint until after the filing timestamp. Verify all details before proceeding.
• Fence the claims for both regimes. Include method + system claims; avoid preference language (“better,” “preferred”). State the technical effect once for the UK (e.g., reduced sensor noise or lower compute at inference) and keep thresholds explicit; if a threshold is still tentative, note it as an example.
• Pre-pack diligence. One-page timeline, docket number, POA copy, and a short known-risks note (e.g., <12-month poster). It signals control and keeps investor questions short.

Next action: open a new provisional draft titled “Signal-guided metabolic feedback—systems and methods,” sketch the six-step pipeline with concrete thresholds, and send it to counsel before close of business today. You’ve done the hard part—choosing a path.

🔗 Three-Day Foreign Filing License (FFL) Posted 2025-10-04 13:40 UTC

Intro: your fast lane today

In the US, writing “recommend” reads as abstraction under §101; in the UK, “feels better” fails the technical-contribution test. Two offices, two filters (think of them as two checkpoints) share the same cure: concrete cause and effect.

We’ll draft two independent claims—one that acts on the body, one that measurably improves the computer—and you’ll get a one-page Data Card and a short consult form. Note: In February 2024, the USPTO said AI-assisted inventions still need a human’s “significant contribution,” and early 2025 UK practice again centered “technical contribution” for AI claims. You’re not alone; we’ll take it one steady step at a time.

• Lane A: Acts on the body Measure twice, cut once—write actions, not opinions, in the claim body: detect → filter → normalize → classify → actuate. Example: “If noise <5% and feature X ≥0.7, deliver 2 mA for 120 s”; skip statements about “recommend,” “favor,” or user preference.
• Lane B: Improves the computer Name the bottleneck and the metric. Example: “Reorder-buffer compaction reduces L2 misses ≥15%”—therefore fewer cache stalls at the same workload—or “streaming quantizer cuts batch-1 latency 8–12%,” therefore tighter tail latency. Tie the gain to the architecture, not taste—no vibes-based scoring.
• Data Card + consult Log inputs, thresholds, units, dataset, and the source of each number (lab note, log); proceed only with values your records support. Then select your primary lane and add the other later as a continuation if needed.

Next step (15 minutes): fill the Data Card and sketch both claims with real numbers so neither office keeps you waiting.

Why this feels hard (and how to simplify)

Personalized nutrition straddles two worlds—biology and software—like standing in a doorway with one foot in each room. If your claim sounds like “analyze data and recommend a meal,” it reads abstract in the US and non-technical in the UK: The key question becomes: what, exactly, are you trying to prove?

Add a concrete dosing step and US prospects improve. Show a measurable improvement in computer operation and UK prospects improve—the same invention, different doors.

The core simplifier is this: Split the story into two lanes: (1) an applied intervention that changes a physiological marker; (2) a technical contribution that makes the system itself work better. Each lane can anchor an independent claim for its jurisdiction.

• Lane A—Body: State dose and timing with an endpoint: “deliver 50 mg magnesium at 21:00 if HRV < baseline−5% for 3 nights; evaluate HbA1c at 90 days.” Avoid verbs like recommend or favor.
• Lane B—Computer: name the bottleneck and metric: “stabilize the sensor pipeline to cut motion-artifact variance below 1% and reduce inference latency by 15%.” This approach clearly shows cause and effect.
• Draft two independent claims. A Vanda-style method-of-treatment claim for the US (2018), and a technical-effect claim for the UK that focuses on signal processing, memory use, or compute—never taste or preference (UKIPO guidance, 2025).
• Keep your proofs tidy—measure twice, cut once. For Lane A, point to the physiological endpoint; for Lane B, point to system-level metrics (e.g., accuracy at fixed latency, energy per inference).

A Micro-Story I once drew a claim ladder on a coffee receipt at 16:40—US “dose–timing” at the top, UK “signal-chain stability” below. That 20-minute pivot cut months of ping-pong.

US: Focus on concrete steps that act on the body and name the endpoint; this aligns with the method-of-treatment path recognized after Vanda (2018).

UK. Focus on technical contribution—improving computer operation or the signal chain—not user preference. That’s consistent with UKIPO’s 2025 emphasis.

Next Action: write one paragraph per lane that names the trigger, action, and measurable result; then lift each paragraph into its own independent claim draft.

The 2025 rules at a glance: US vs UK

When an ML thread runs through a claim, the rules can feel slippery—like stepping onto a wet platform on a rainy morning. Here’s the firm footing for 2025. If you’ve read this far, you’ve already done the hardest bit.

Inventorship (both). Name the human. Tools can assist, but the inventor must be a natural person. In the US, 2024 guidance requires a “significant human contribution.” In the UK, the Supreme Court’s DABUS ruling confirms machines can’t be inventors.

US §101 (subject-matter eligibility). Algorithms or data processing alone risk the “abstract idea” bucket; method-of-treatment claims with concrete dose and timing remain eligible (e.g., Vanda, 2018). Pure diagnostics still face headwinds (e.g., CareDx, 2022). 2024 updates did not change the core test: tie your claim to a practical application and to what the human actually did.

UK technical-contribution test. Start from the “program for a computer as such” exclusion and get through it by showing a technical effect. After the 2023–2025 practice updates around Emotional Perception, a trained ANN isn’t excluded just because it runs on a computer; but outputs that boil down to taste or aesthetics (like music picks) are weak unless you can point to a signal-processing or system-level improvement.

Working rule. Name the human, prove the effect, and make the computer do something it couldn’t do before.

• US—prove intervention. Write the dose and timing, then show an objective change (e.g., “deliver 50 mg at 21:00 if HRV dips 5% for 3 nights; evaluate HbA1c at 90 days”). Avoid verbs like “recommend” without an action step.
• UK—prove technical contribution. Quantify a computer-level gain (latency, memory footprint, stability) or a signal-quality effect (e.g., SNR). One number beats ten adjectives.

Next action. Draft two independent claims: (A) a method-of-treatment with specific dose/time and a measurable endpoint; (B) a computer-implemented method that improves a named metric (e.g., 12% lower inference latency on-device). Keep lab notes linking the named inventor to the decisive step. A small tidy today can spare you a scramble tomorrow.

US §101 playbook (what works)

What, exactly, keeps you on the right side of §101? Anchor to application by framing the model as part of a treatment workflow: sensor signal → biomarker trend → specific dose & timing → lab metric improvement. Think of it as the steady morning commute, not a scenic detour; this places you on the Vanda path (USPTO memo, 2018; Justia, 2018). Avoid “observe and infer” claims that look like diagnostics (CareDx, 2022).

Practical drafting moves—if you’ve read this far, you’ve already done the hardest bit; keep going.

State the analyte/class (e.g., branched-chain amino acid), the dose band (e.g., 0.15–0.30 g/kg), and the timing window (e.g., within 20 minutes post-exercise).

Force objectivity: tie to a lab endpoint (HbA1c, LDL-C) or a physiological metric (HRV SDNN).

Place the AI in-line: “classifying photoplethysmography artifacts to maintain signal SNR ≥ 20 dB before calculating trend,” then act.

Micro-story. A founder pivoted one verb—from “recommend” to “administer”—and added a 90-day HbA1c endpoint. The §101 rejection vanished; the examiner’s next question was dosage range support. A small tidy today can spare you a scramble tomorrow.

Show me the nerdy details

US evidence pointers: Vanda (2018) upheld claims because they applied a natural correlation via dosing adjustments. CareDx (2022) invalidated diagnostic claims that measured cell-free DNA but stopped short of a treatment step (Justia, 2018; USPTO, 2024-07).

UK technical-contribution playbook

Open with how the machine works better. Name the bottleneck, then the measurable lift: tighter cache locality (keep data near the CPU), fewer memory copies, motion-robust photoplethysmography, or sensor fusion that halves jitter—like keeping the kettle near the tap, not across the kitchen. That is the gate the UK uses—“technical contribution,” not taste or training bravado.

Recent practice reinforces this. The 2025 UKIPO AI guidance centers technical effect; and in 2024 the Court of Appeal held an ANN-based music-recommendation system was still a “program for a computer as such,” so outputs like song picks carry little technical weight on their own. Anchor your claim to a concrete improvement in computation or signal handling.

• System claim. Wearable + embedded model + fixed-point pipeline (integer math) that cuts energy by ≥15% at equal accuracy. Spell out the pipeline stages you removed or fused to get there.
• Compute claim. On-device 8-bit quantization with dynamic-range clipping that holds median latency ≤10 ms for HRV windowing (e.g., 60–120 s windows). State the buffer size and cache hits you improved.
• Signal claim. Artifact rejection that lifts SNR from 14 dB to 23 dB while jogging at 160 bpm. Name the filter (e.g., adaptive notch + accelerometer gate) and the test protocol.

One gentle reminder: skip “better recommendations.” Give the examiner a number—crisp, reproducible.

Next action: Write one sentence that ties a specific change to a measurable effect (e.g., “merged resampler + detrend step reduced memory copies by 2→1 and dropped power 18% at 50 Hz”). Then map it to the system, compute, or signal claim above. A small tidy today can spare a scramble at examination. ::contentReference[oaicite:0]{index=0}

Show me the nerdy details

Apply Aerotel/HTC signposts: does the claim control a technical process outside the computer, improve the computer itself, or introduce a new operating mode? If yes, you’re safer. Highlight quantitative compute/signal gains; omit “subjective wellbeing” as the key effect (GOV.UK, 2025-01).

EU/UK treatment & diagnosis exclusions (and workarounds)

If you’re filing in Europe or the UK, the medical exclusions can feel like a moving target—like catching a train on a rainy morning. So how do you keep your footing?

The Rule Under EPC Art. 53(c), methods of treatment of the human or animal body and diagnostic methods practiced on the body are excluded. Products, devices, and purpose-limited medical uses remain patentable, including “second medical use” claims.

Useful Nuance The diagnostic exclusion usually bites only when a claim includes all steps—measurement, comparison, finding a deviation, and the diagnostic decision—and at least one step is performed on the body. Keep steps ex vivo/in vitro and avoid claiming the final diagnostic decision to lower risk. If you’ve read this far, you’ve already done the careful work examiners respect.

Wellness Lane If the project is truly non-therapeutic, say so explicitly. Frame outcomes as performance or general wellbeing and keep claims ex vivo/in vitro or device-centric; that reduces Article 53(c) exposure and keeps downstream marketing cleaner.

Therapeutic: “Composition X for use in lowering LDL-C in adults with baseline ≥160 mg/dL.”

Wellness: “System for adjusting micronutrient timing to stabilize HRV during non-clinical training.”

• Pick the Right Claim Type For treatment, use a product claim “for use in” a therapeutic purpose; for wellness, prefer device/system claims that act on data or schedules, not the body.
• Keep the Body Out of the Steps Place calibration and analysis in vitro or on recorded data; avoid claiming on-body intervention or a diagnostic conclusion.
• Be Explicit About Intent Use plain language—“non-therapeutic,” “performance,” “training”—when that’s accurate. If symptoms are in scope, examiners may read it as therapeutic.
• Separate Lanes in Practice Run clinical validation under regulatory counsel; keep the patent story to devices, data handling, and measurable non-clinical outcomes.

Micro-Story A client pitched a “sleep aid” and drifted into therapy-claim territory; we recast as “device + in-vitro calibration + user-facing schedule generator,” and clearance followed in 12 days while the clinical work continued under separate counsel.

Next Action: Write one sentence that states your lane (“therapeutic” or “wellness”) and where each claimed step occurs (in vitro, ex vivo, or on data). Then draft to that map. A small tidy today can spare you a scramble tomorrow.

Sufficiency: data, training, and plausibility (G 2/21)

Worried an examiner will say “interesting, but not reproducible”? You’re reading the room: in 2025, the EPO expects in-spec detail that lets a skilled person retrain—or at least see the necessary data traits. Post-filing evidence can support the story only when the effect was already plausible on the filing date (G 2/21). And as T 161/18 shows, vague training disclosures invite trouble—so keep it concise and organized: name your sources, preprocessing, distributions, and labels.

• Provenance & Rights State dataset origin (public/private), the consent/licensing basis, and any data-use limits. If you mixed sets, give the ratio and the date ranges.
• Traits & Hygiene Give inclusion/exclusion rules, class balance (e.g., 62:38), and known artifacts (e.g., motion blur; duplicate frames). Say how you filtered or deliberately retained them.
• Training Recipe List preprocessing (resize/normalize), augmentations, loss, optimizer, batch size, LR schedule, and the early-stop rule. Mention label-quality checks and who/what performed them.
• Leakage & Validation Explain split logic (subject-level, time-based), leakage guards, and the primary metric with confidence intervals. Add one simple baseline and say why it’s fair.
• Repro & Compute Provide seed ranges, key hyperparameters, and a budget (e.g., 12-hour train on T4; 8-bit inference ≤10 ms on CPU). Note any hardware-specific kernels.

Next action: draft a 2-page “Data Card” in that order, plus a one-pager comparing baselines vs your model. Keep numbers small and verifiable; careful documentation today can prevent problems tomorrow.

Show me the nerdy details

Drafting scaffolds: (i) name sensor types and rates; (ii) specify augmentation bounds; (iii) report ROC-AUC with CI; (iv) report latency/energy under realistic loads; (v) show one comparative table (SOTA vs you). Cite the principle, not vendor names.

Claim-set templates you can adapt

US independent method—treatment. “Obtain sensor signals from a wearable; compute a biomarker trend using artifact-rejected windows; administer a dose of [amino acid/micronutrient] within [X] minutes after [trigger] when the trend exceeds [threshold]; thereby reducing [HbA1c/LDL-C] by ≥[Δ] over [N] days.” Frame this as a true method-of-treatment (MOT) claim—keep an active dosing step, not a mere recommendation; think of it as a streamlined process from signal to action.

US independent system: “A wearable and an on-device model with fixed-point inference achieving ≤10 ms latency while maintaining accuracy ≥AUC 0.85; a controller that triggers dispensing or schedule adjustment when SNR ≥20 dB.” Pair latency with accuracy, and tie the controller action to a measurable signal condition—keep it concrete.

UK/EP independent system/computer-implemented: “A sensor-processing pipeline comprising artifact detection, constrained Kalman smoothing, and quantized inference that reduces median energy per analysis by ≥15% at equal accuracy; outputting instruction sequences for nutrition scheduling.” Keep the technical contribution upfront—compute or signal-handling gains—before any lifestyle outcome, and you’ll stay on firmer ground.

Next action: paste these three shells into your draft and replace the bracketed fields with your biomarker, Δ value, threshold, and timing window (e.g., 20–30 minutes post-trigger)—if you’ve read this far, you’ve already done the hardest bit.

Show me the nerdy details

Layer dependent claims: (i) specify window sizes; (ii) define thresholding math; (iii) claim on-device caching policy; (iv) bind to measurable deltas (ms, mW, dB). In EP, shift “treatment” verbs into product claims or medical-use formats to dodge 53(c) (EPO Guidelines, 2024).

Comparison tables (US vs UK; Wellness vs Treatment)

US vs UK claim strategy stack (2025)

Layer	US (AI-driven personalized nutrition)	UK (AI-driven personalized nutrition)
Independent claim verb	Administer / “apply dose & timing” (Vanda)	Process/Control / “reduce latency/energy, stabilize signal”
Objective metric	HbA1c, LDL-C, HRV SDNN over N days	Latency ≤10 ms; energy −15–25%; SNR ≥20 dB
Primary risk	Abstract idea if “recommend only” (CareDx risk)	No “technical contribution” if subjective output
Sample clause	“…administering 0.20–0.28 g/kg within 20 minutes post-exercise…”	“…quantized inference with jitter RMS ↓ ≥22% at 160 bpm jog…”
Keyword to sprinkle	“AI-assisted inventorship guidance 2024”, “CareDx diagnostics §101 risk”	“UKIPO 2025 AI guidelines technical contribution”, “ANN as computer program Emotional Perception 2024”

Wellness vs Treatment framing (EP/UK Art. 53(c) context)

Framing	Allowed focus	Risk	Alternatives
Wellness	Performance, general wellbeing; ex vivo/in vitro processing; device/system	Marketing creep into clinical claims	Keep outputs non-therapeutic; anchor to compute/signal effects
Treatment	Substance/composition “for use in…”; device/system implementations	Method-of-treatment/diagnosis exclusions	Use medical-use format; avoid in-body diagnostic methods

Conversion hub: consult + free Data Card template

Two paths for next steps. Both keep a safe 160 px distance from any ad unit for policy and UX sanity.

Protecting datasets & rules without tripping exclusions

UK practice in 2025 treats datasets “as such” as non-patentable information. When a dataset functions as a component that delivers a technical contribution—say, a calibration table that cuts jitter or a lookup curve that stabilizes inference—it can be claimed as part of the system that achieves that effect (like switching on a small desk lamp so the work stays steady).

If you’re balancing disclosure and defensibility, a layered plan keeps you safe and practical. What exactly are you trying to protect: the table itself or what it does? If you’ve read this far, you’re doing the careful work that pays off.

• Patent the technical use, not the content. Claim the dataset as a module in a signal/computation pipeline with a measurable lift (e.g., “reduces false positives by approximately 50% at equal sensitivity”); use “technical contribution” (aka technical effect) language and name the metric and test setup.
• Maintain rules as trade secrets. Place labeling rubrics, redaction heuristics, and decision thresholds in a private standard operating procedure (SOP) with access controls and change logs; publishing those rules risks destroying secrecy, while referencing the outcome in claims does not.
• Utilize copyright and database rights where applicable. Protect selection/arrangement and any original annotations; consider the UK and EU sui generis database right for substantial investment in obtaining/validating the compilation.
• Document provenance and effect. Record dataset sources, curation dates, and before/after performance so your claim chart and trade-secret file are audit-ready.

Case Study: A team once tried to patent a recipe list—“if HRV drops, eat X.” We rewrote around the artifact-rejection lookup that actually halved false positives and moved the dietary “rules” into a locked SOP; cleaner claim, stronger moat.

Recommended next steps: draft a system claim that names the lookup/calibration dataset as a functional module tied to a specific performance gain, and move your labeling/redaction rules into a version-controlled SOP with restricted access. Careful preparation today can prevent complications tomorrow.

Governance: inventors, logs, and AI use disclosure

If you’re juggling filings, demos, and shifting teams, a missed name can derail an otherwise clean application. Treat this as a quiet pre-flight check before you head out the door.

People first. Name the human inventors who made a significant contribution, then keep a living contribution matrix—design, experiments, thresholds, evaluation—with initials, dates, and a proof line (commit hash or PR). What, exactly, slips when we move fast? Recent USPTO guidance (2024) expects clear human input. If AI tools helped, note the model, version, and what it did. Keep that record internal unless and until disclosure is appropriate.

Contracts & timing. Verify assignment agreements for employees, contractors, and collaborators before merges. Freeze public demos and talks until a priority filing is secured. For diligence, maintain one PDF: versioned figures plus a one-page “why it works” note that ties mechanism to data. If you’re already doing part of this, keep going—steady wins here.

• Stand up the matrix today: four columns—who, what, when, proof. Example: “J.K. — artifact filter threshold 0.35 — 2025-09-12 — PR #184.”
• Centralize evidence: export figures to PDF; use filenames like 2025-10-09_fig3_v2.pdf; link back to raw data.
• Lock assignments: confirm work-for-hire language for contractors and store signed copies with payment receipts.
• Control disclosure: keep a green/red list for posts and talks; move items to green only after filing.

Case Study A contractor’s tweak to an artifact filter was almost overlooked; a late code review caught it. Adding one name early was cheaper than a three-month inventorship correction.

Next action: open a “Contribution Matrix” doc, add all contributors from the last 90 days, and pause any public demo until the filing date is on the calendar. A small tidy today can spare you a scramble tomorrow.

Filing sequence & geo strategy (US+UK play)

If the timeline feels uncertain, think of this plan as switching on a small desk lamp—just enough light to move without overcommitting. Build one shared core—sensors, features, pipeline—and let the claims branch cleanly by jurisdiction. Steady steps. No drama.

Two-track independence: From the same spec, draft two independent claims: (1) US: a true treatment action with an objective endpoint; (2) UK: a concrete technical contribution that measurably improves computation or signal handling. Shared facts; different gates. If you’ve read this far, you’ve already done the hardest bit.

• Day 0: File a US provisional or GB priority with a full Data Card and baseline metrics. Name sources, preprocessing, and test setup.
• Day 30–90: Prepare the PCT (Patent Cooperation Treaty) draft with dual claim ladders; aim to file by the 12-month priority deadline. Use the international phase to pilot a UK-leaning claim set against the 2025 AI signposts.
• Month 18: Expect PCT publication; adjust examples and comparative data while you still have room to refine.
• Month 30–31: Enter national phase in US and EP (and GB if needed). Keep the US method-of-treatment independent; keep the UK/EP independent anchored to a technical effect on compute/signal.

Cadence inside the company: GB-priority → PCT decision cycles often run 10–15 days in practice. In the US, Office Action responses are typically due in 3 months (extendable to 6); plan experiments and rebuttal evidence to fit that window. Quiet prep now saves rush later.

Investor-ready signal: Pair one clinical endpoint (e.g., −0.4% HbA1c at 90 days) with one technical endpoint (e.g., −18% energy per analysis at equal AUC). Two clear graphs beat twenty adjectives—and travel well across boardrooms.

Next action: Open the Data Card today: list sensors, features, baselines, and the 90-day clinical and technical endpoints you will commit to measure. A small tidy today can spare tomorrow’s scramble.

Show me the nerdy details

Label figures with region tags: “(US) HbA1c outcome,” “(UK) latency/energy,” “(EP) device/system wiring.” Keep marketing claims aligned: wellness vs therapy lines must match your claim framing (EPO Guidelines, 2024; GOV.UK, 2025-01).

Local notes: timing, cost, and process (US · UK)

United States (2025). Expect the first non-final Office Action on a typical software/AI application within a broad range depending on art unit; what you control is the quality of the first response. Budget planning many startups use (illustrative): $12k–$22k for drafting and prosecution through a first OA; PCT national stage US fees vary by entity size (USPTO fee schedule, 2024—data here moves slowly). Response windows are generally 3 months standard, extendable to 6 months.

United Kingdom (2025). GB priority filings are attractive for quick starts; many teams slot internal decision gates at 10–15 days for GB → PCT. Expect UKIPO to probe for “technical contribution.” Typical advisory budget ranges: £8k–£18k through first substantive examination for compute/signal-heavy claims. Keep the Emotional Perception lesson in view: avoid subjective goals; talk ms, mW, and dB.

Humane aside. Draft near a window if you can; a quiet morning helps you shave 10% off word count and 30% off ambiguity. Numbers not adjectives—that’s your compass.

AdSense UX & policy stability (2025-safe)

Spacing. Keep heavy-click UI (download buttons, forms) at least 160 px away from ad slots to reduce mis-clicks and policy risk. You saw that in this article.

Height reservation (CLS). Reserve space so layouts don’t jump. Use a wrapper around your ad with a fixed min-height that matches your typical in-content unit.

<div aria-label="ad-slot-wrapper" style="min-height:280px;margin:24px 0;"> <!-- your Adsense <ins> goes here --> </div>

💡 Read the EPO AI & ML examination guidance

FAQ

Can I name my AI tool as an inventor?

No. The US requires a human with a significant contribution (AI-assisted inventorship guidance 2024; Federal Register, 2024-02). The UK Supreme Court held that an inventor must be a natural person (White & Case LLP, 2023-12).

We only “recommend” foods. Is that patentable?

US: risky—pure recommendations look abstract unless tied to a concrete treatment step and objective endpoint (CareDx diagnostics §101 risk, 2022; USPTO, 2024-07). UK: risky unless your computer works better (e.g., latency/energy/SNR) (GOV.UK, 2025-01).

What about diagnostics using biomarkers?

US diagnostics remain tough; consider adding a treatment action or pursue device/system claims (CareDx, 2022). In EP/UK, avoid in-body diagnostic method claims; keep ex vivo/in vitro or device-centric (EPO Guidelines, 2024—data here moves slowly).

Can I patent my training dataset?

In the UK, a dataset “as such” is excluded; but a dataset that enables a technical contribution within a system can be claimed as part of that system (GOV.UK, 2025-01).

How much training detail must I disclose?

Enough to make the effect plausible and reproducible: sources, preprocessing, label definitions, distributions, and baseline comparisons (EPO G 2/21 plausibility data card, 2023; EPO Case Law, 2022).

How do I avoid an ANN being treated as a “program” in the UK?

Show a technical contribution outside the “program as such” exclusion—e.g., stabilized sensor pipeline, reduced latency/energy, or improved memory behavior (UKIPO 2025 AI guidelines technical contribution; GOV.UK, 2025-01).

What’s the rough cost and timeline to tie US and UK via PCT?

Founders commonly set aside a five-figure USD/GBP budget for drafting + early prosecution, plus PCT fees at month 12; internal decision lead time GB→PCT often 10–15 days; OA response windows are typically 3 months in the US (extendable). These are broad, planning-level ranges observed in 2024–2025.

Patent vs trade secret for my rules?

Patent the technical use (how the rules improve compute/signal); reserve the rulebook specifics as a trade secret. Publish the mechanism, not your exact thresholds.

Conclusion & 15-minute next step

If this still feels slippery, that’s normal. We’ll pin it down in one short pass—like setting the kettle on and clearing the counter: quick, deliberate—and if you’re here, you’ve already done the hardest part.

Same invention, two filters. the US claim acts on the body and proves it; the UK claim improves the machine and proves that—a technical contribution (often called a technical effect). Two doors, one corridor. What’s the move now? We won’t add new data here; we’re just framing what you already have.

1. Create the file. Title it 2025-AI-nutrition-dual-claims-v1.
2. Add two headings. “(US) Treatment Action + Endpoint” and “(UK) Technical Contribution.”
3. Write three lines total.
   • US: one dose+timing line (e.g., “Deliver [dose] at [time/trigger]”) and one objective endpoint (e.g., “evaluate [metric] at [N] days”). Keep it crisp—no extra prose.
   • UK: one pipeline improvement line (e.g., “latency -12 ms” or “energy -15% at equal accuracy” or “SNR ≥ 20 dB after artifact rejection”). That’s your latency budget in one stroke.
4. Download the CSV template and fill 8 rows: 4 for US (dose, timing, endpoint, test window) and 4 for UK (bottleneck, change, measurement, proof note). Email it to counsel with subject Dual-claims v1 — review.

Short story. A drizzly London morning, 07:30 in Soho. A founder slid over a laptop: “Our model finds calmer playlists after oatmeal.” We circled “calmer,” wrote “motion-artifact jitter ↓ 22% (RMS) at the photodiode”—therefore a steadier signal to start with—and added a cache-friendly buffer note that trimmed ~ 12 ms. Same code, reframed. Two weeks later, their draft read cleaner—and moved.

Do this now: open a blank doc, paste the two headings, add the three lines, complete 8 CSV rows, and hit send. If you pause mid-way, leave the file open and finish the rows today— a small tidy now avoids a scramble tomorrow.

---

US Lane

• Hook: Method-of-treatment with dose & timing.
• Metric: Objective endpoint (HbA1c, LDL-C, HRV).
• Risk: Pure diagnostics (CareDx, 2022).
• Support: Vanda (2018); USPTO AI §101 update (2024-07).

UK Lane

• Hook: Technical contribution (compute/signal).
• Metric: Latency/energy/SNR improvements.
• Risk: Subjective outputs (Emotional Perception, 2024).
• Support: UKIPO AI guidelines (2025-01).

A founder’s cheat-sheet: same spec, different doors. Numbers beat adjectives in both lanes.

Evidence cues used: (Federal Register, 2024-02); (USPTO, 2024-07); (White & Case LLP, 2023-12); (GOV.UK, 2025-01); (EPO Guidelines, 2024/2025); (Justia, 2018); (Reuters, 2024-04).

Last updated: 2025-10-09 (KST)

US vs UK patentability of AI-driven personalized nutrition, AI nutrition patents, Vanda method-of-treatment eligibility, UKIPO 2025 technical contribution, EPO G 2/21 plausibility

🔗 Trademark Registration Posted 2025-09-29 03:45 UTC 🔗 HFT-Related Patents Posted 2025-09-27 03:18 UTC 🔗 Provisional Patent Guide Posted 2025-09-25 08:09 UTC 🔗 Space Tourism CPC Codes Posted 2025-09-21 UTC