Limited time offer: try Patentext for free →
.svg)
If you've landed on this page, you're probably researching AI tools to help streamline your patent drafting. Maybe you’ve come across DeepIP — it’s one of the more visible players in the space. But like many tools built for enterprise, its website can feel opaque: it’s not immediately clear how the product works, who it's for, or whether it’s worth scheduling a demo just to find out.
Another name you might have seen is Patentext. Our tool is newer, smaller, and built from the ground up by patent practitioners. But how exactly is it different from DeepIP? And is it a viable contender if you're evaluating tools for serious drafting work?
This guide cuts through the noise. By the end, you’ll have a clearer sense of which fits your drafting process, quality standards, and review habits.
DeepIP began as Davinci, an AI-powered patent assistant that quietly entered beta in early 2024. By August of that year, it had rebranded to DeepIP and had officially launched, positioning itself as an “AI co-pilot” for patent professionals, primarily running inside Microsoft Word.
The company was founded by François‑Xavier Leduc and Édouard d’Archimbaud, both of whom come from deep AI and data infrastructure backgrounds. Before DeepIP, they co-founded a platform for building secure, enterprise-grade data pipelines for machine learning.
This technical foundation shapes much of DeepIP’s philosophy: it’s a tool built by AI experts to modernize an industry the founders saw as lagging in automation.
Patentext is an AI patent drafting platform that takes a different approach from most tools in the space. Instead of relying on open-ended chat prompts or document uploads, it uses a structured, visual framework to model the invention. That framework serves as the foundation for generating each part of the application, including the claims, specification, and abstract.
Although Patentext officially launched in March 2025, it’s grounded in over a decade of hands-on patent drafting experience. Our team has worked at both large law firms, like Fenwick & West, and smaller boutique practices alike.
This experience drafting for early-stage startups, unicorn tech companies, and lean legal teams has shaped how Patentext approaches structure, consistency, and usability.
DeepIP is a chat-style assistant embedded directly into Microsoft Word, with an additional web-based option. Users interact with it by typing natural language prompts to request tasks like drafting claims, summarizing disclosures, describing figures, or generating responses to office actions.
Under the hood, DeepIP uses retrieval-augmented generation (RAG) techniques alongside large language models trained on public and proprietary patent datasets. The tool is designed to make patent drafting feel more accessible, especially for those familiar with GPT-style interfaces.
Patentext takes a fundamentally different approach. Once users upload a disclosure (e.g., notes, a scientific paper, a transcript), the AI creates a visual graph that maps the invention’s steps, features, and relationships.
The graph structure does two things at once: it captures the logic of the invention in a way that’s editable and auditable, and it gives the AI a clear context to work from. Rather than guessing at what to say or how to structure a section, the model pulls directly from this graph.
DeepIP’s chat-based interface makes it easy to iterate on drafts. Ask the AI to rephrase a claim, summarize a disclosure, or rewrite a section, and it delivers within seconds.
But because the system relies on prompt-driven generation, the quality and tone of the output can vary significantly depending on how a request is phrased. As one DeepIP customer noted in a recent webinar, “It’s like having an army of interns. If you ask the intern a very vague question, you’re probably going to get an unfocused response, or worse, gibberish.”
For some teams, particularly those already comfortable working with generative AI tools, reviewing and refining the results is part of the process. But for teams prioritizing consistency across longer or multi-author drafts, the variability can create extra editing work.
Patentext approaches draft generation differently. By using a shared invention graph to power all sections of the application, the platform ensures that language choices — like how a component is described — remain consistent throughout. This reduces the effort required to align tone and terminology across sections.
However, Patentext’s focus today is more on logical consistency than stylistic nuance. The platform doesn’t yet allow users to define or replicate personal drafting style, though we’re actively developing new features to support customizable language and tone.
For now, teams that value clarity, traceability, and internal alignment may find the structure beneficial, while those seeking more tailored prose might still prefer to finesse the final language manually.
DeepIP is designed to integrate seamlessly into existing workflows. Its ribbon-style add-in brings the AI directly into MS Word’s interface, reducing the need to learn new tools or platforms.
That said, in the same webinar, the customer described their onboarding as an “incremental” process: uploading disclosures, adding drawings, building claims, and gradually developing a workflow around the tool. In their words, DeepIP starts to deliver real value “once you have an ecosystem of context in place.” Until then, users may need to experiment and refine their approach to get consistent results.
Patentext takes a more guided approach from the start. Because the workflow is clearly defined and doesn’t require prompt engineering, onboarding is often more straightforward, particularly for practitioners who are used to thinking in terms of invention logic rather than natural language commands.
That said, because Patentext doesn’t follow a familiar chat interface or document-based editing model, it can feel less immediately intuitive for users expecting a blank canvas. You’ll need to spend some time understanding the different elements Patentext uses (e.g., a node, an attribute, a component) to get the most value out of the platform.
For teams that prioritize consistency and want a more transparent drafting process, this tradeoff can be worthwhile. Those seeking faster experimentation or more freeform interaction, however, may find the graph-based approach more structured than they’re used to.
DeepIP integrates cleanly into existing Word-based drafting and review workflows. You can generate sections right in Microsoft Word (or use the web app) and share files for review or comment, just like any other DOCX. It also includes version control and audit trail features, capturing changes and keeping a record of who did what.
However, collaboration occurs primarily at the document level. There’s no visibility into which prompt generated which section and no in-tool functionality for inline AI-comment tracking or multi-user editing. Reviewers often need to rely on manual notes to understand draft origins, and recreating or modifying content may require prompt retries rather than structured edits.
For teams with established review processes, this may not be a blocker, but it can add friction when multiple people are trying to coordinate feedback or edits within the AI workflow.
Patentext doesn’t yet support multiple users editing the same draft in real time, but it does have foundational collaboration tools in place. Each team member can work on their own invention graph, while account administrators can view all drafts created within the organization.
And because each paragraph in a draft maps back to a specific node in the graph, reviewers can quickly understand the logic behind a given phrase or section. This makes it easier to review progress and maintain oversight, even without direct editing access.
We’re hoping to introduce full multi-user collaboration soon, including shared editing sessions, inline comments tied to the underlying graph, and version histories structured around the invention model.
One of DeepIP’s latest features is its Office Action module, which allows users to upload examiner communications and receive AI-generated support. The tool highlights key objections and proposes draft response language, streamlining what can often be a time-consuming part of prosecution.
This can be especially useful for firms looking to reduce attorney time on initial OA analysis or generate quick response templates for review.
Patentext recently introduced a References feature aimed at reducing the manual effort involved in managing terminology and figure references across a draft. Instead of updating terms or numbers section by section, users can:
This feature helps reduce the small but error-prone tasks that often slow down finalization, particularly when terminology shifts late in the drafting process.
DeepIP does not publish pricing on its website. Firms interested in using the tool need to request a demo or contact sales for a quote — unsurprising, given their pricing model is more tailored to enterprise or high-volume use cases.
Patentext’s early access plan is priced at $200/month. This includes full access to the drafting platform, invention graph, and generation tools for claims, spec, and abstract, along with Word export and direct support. There are no seat minimums or onboarding fees, making it easier for solo agents, boutique firms, and lean legal teams to get started.
Note: Early access pricing ends soon! Lock it in before we do our full release.
DeepIP fits well into larger organizations or teams with high-volume drafting needs and existing document-based workflows.
Per the webinar we’ve been mentioning, one customer described how they’ve had to develop internal best practices to get the most out of DeepIP, including prompt training, staff-led demos, and workflows that paired generated drafts with prompt transcripts for better review.
“Everybody learns from each other,” the speaker noted, emphasizing the collaborative effort required to make the tool work consistently at scale.
For teams willing to invest in training and internal systems, DeepIP can be a powerful drafting assistant, especially when speed and enterprise-grade compliance features are top priorities.
Patentext is built for lean teams who draft patent applications themselves and care deeply about clarity, auditability, and consistency, such as:
Both DeepIP and Patentext are tackling the same problem of AI-assisted patent drafting, but we’re solving it in fundamentally different ways.
DeepIP focuses on speed and accessibility, especially for teams already working in Word and familiar with chat-based tools. It offers flexibility, fast iteration, and strong enterprise integrations. But it also asks teams to bring their own structure to get consistently high-quality output.
Patentext, on the other hand, builds structure from the start. Its graph-based model gives the AI clear context and lets practitioners drive the drafting logic without requiring dozens of back-and-forth prompts. That can make it easier to produce clean, consistent drafts quickly, but the tradeoff is a slightly steeper learning curve.
In short:
Curious how Patentext works in practice? Take our self-guided product tour or schedule a live walkthrough with a sample disclosure.
