Habr AI compares AI rewriters: many services turn out to be ChatGPT wrappers

The market for AI rewriting for Russian-language newsrooms turns out to be much less technologically advanced than its packaging suggests. A large review on Habr AI shows that a significant share of services either still solve the old SEO problem of raising a uniqueness percentage for a report, or simply sell access to the same LLMs through a more convenient interface at a noticeable markup. Real editorial quality today comes not from "magic buttons" but from well-configured models, prompt libraries and multi-agent pipelines with several stages of text checking.

The review covers 24 tools, and the author divides them into four broad categories. The first is old-school SEO rewriters such as Text.ru, Advego, Raskruty and ETXT.

These services mostly shuffle words and pick synonyms to reach the required uniqueness percentage, but barely work with meaning. The second category is general-purpose LLMs: Claude 4.6 and 4.

7, GPT-5, Gemini 3, DeepSeek V3, YandexGPT 5 and GigaChat. They are noticeably stronger on text quality, but without a good prompt they still often produce an averaged, "chatbot-style" result. The third group is ready-made wrappers built on the same models, for example PR-CY, Turbotext, Chad, Gerwin AI and NeuralWriter.

Their value usually lies in the interface, templates and billing rather than in proprietary technology. Finally, the fourth category is multi-agent and vertical pipelines, where several models or agents take turns gathering facts, writing a draft, critiquing it and polishing it into a final version. To make the comparison more than theoretical, the author ran one short news item about agents, around 1,500 characters long, through several services with the simple request "rewrite this text in your own words."

In this test Claude Sonnet 4.6 got the best subjective score, 7.5 out of 10, with minimal clichés and zero hallucinations.

GPT-5 landed slightly lower at 7 out of 10. DeepSeek V3 looked strong on price-to-quality ratio but made a factual error and inserted an extra number. YandexGPT 5 and PR-CY delivered average results, while the old Raskruty synonymizer essentially just shuffled words and scored 2 out of 10.

The review also stresses that no product in the selection offers full fact-checking out of the box at the level of a human editor, so human verification is still mandatory. The review also points to a practical fork depending on use case. If a newsroom needs to rewrite one or two articles a month, it is usually cheaper to go straight to Claude or ChatGPT and spend time once on a decent prompt.

Once volume reaches 10–30 pieces a month, in-house prompt libraries in Notion or Word start to pay off, as do convenient LLM wrappers. If the goal is to consistently produce 30 or more texts in the style of a specific publication, with background context, fact-checking and draft refinement, multi-agent solutions come to the fore. They are pricier and more complex, but editorial quality today is built around the process, not around a single model.

For companies that must keep data within Russia and comply with the requirements of Federal Law 152-FZ, YandexGPT and GigaChat may be a separate argument, although they trail Western models in quality and in freedom to handle news topics. The central thesis of the piece is that in 2026 the word "rewrite" covers several different tasks at once. It can mean substantive reworking of a news item to a particular outlet's style, adapting the same story for Telegram, Dzen or a longread, or mechanical uniquification of text for SEO checks.

When a client lumps these tasks together, they almost inevitably overpay or get a poor text. The author explicitly notes that the comparison scale is subjective and based on a single test text; in addition, the author's own product is mentioned in the multi-agent category. The review is therefore useful not as a definitive ranking of winners but as a map of the market: old SEO tools still survive on uniqueness requirements, direct access to strong LLMs remains the most flexible option, and real differentiation begins where an editorial process is layered on top of a model.

For media outlets this means one simple thing: paying for an "AI rewriter" only makes sense when the service adds real value, such as style control, a convenient workflow, data restrictions or multi-step text processing. If the service is just ChatGPT or another mass-market model with no serious layer on top, direct access to an LLM will almost always be cheaper and more flexible. The market's bottleneck is no longer text generation itself but how to turn a draft into reliable material that a newsroom can publish under its own name without embarrassment.