The advertisement was unusable. In 2019, Yaniv Makover asked an early fine-tuned language model to write an ad for Toyota. It came back with a profane instruction to abandon Mazda and try Toyota instead. No tasteful photograph of a winding mountain road could rescue this particular campaign. A marketing department would have sent it back. Makover saw a reason to keep going.
The model had connected a brand, a rival and the purpose of an advertisement. Its manners required attention. Its grasp of context suggested something worth developing. For an engineer interested in language, that was enough to make a failed draft interesting. The gap between what a machine could produce and what a customer could use would become familiar territory.
Today, as Anyword’s co-founder and CEO, Makover works on a question that survives even when the prose improves: how do you choose what to publish? Give someone a thousand plausible sentences and the blank page has disappeared. The responsibility for selecting a sentence has grown. His career offers a view of that problem from both sides of the screen.
A long apprenticeship in words
Makover says he began coding at seven. Software appealed to him because it offered a way to build things. Language supplied a particular fascination: the possibility that a machine might participate in a conversation. Early chatbots were limited, but their attempts to imitate human exchange held his attention. Fluency was a distant prospect. The attraction was already there.
At Ben-Gurion University of the Negev, he studied computer science and later information systems, data mining and software engineering. His background includes research in natural language processing. A 2007 paper lists him alongside Menahem Friedman, Mark Last and Abraham Kandel, examining methods for detecting anomalies in web documents. It is a quieter entry on a career timeline than a funding announcement, and a useful one.
Document analysis asks software to find structure in collections of text. Marketing eventually asks a related, more commercially impatient question: what will a reader do with these words? The movement between those questions helps explain the continuity in his work. Language remained the material. The setting changed.
Before his own companies, Makover worked in engineering and technical leadership at Nokia Siemens Networks and Siklu Communications. Building a product involves more than having an idea about an algorithm. Teams must make it run, keep it useful and agree on what comes next. That experience belongs in his story alongside the research.
“I’ve been coding from age 7.”
Yaniv Makover, 2023
The reader at the other end
Keywee, the business Makover says he co-founded in 2013, brought him into publishing. It helped organizations including The New York Times and The Washington Post acquire audiences. The reader might encounter an article through a paid social post rather than by opening a newspaper or visiting its homepage. A story needed a route into someone’s attention.
That placed words inside a system of distribution. An article had a subject; a publisher had an audience; an advertisement had a job to do. Putting the three together meant taking the reader’s likely interests seriously. A piece of writing could be perfectly competent and still reach the wrong person, at the wrong time, with the wrong invitation.
The business relationships were made by people. Makover has recalled the effort involved in reaching American publishers from Israel. One connection became especially durable: his New York Times contact later helped him choose a wedding ring for his wife. There are customer relationships, and then there are customer relationships with opinions about jewelry.
The anecdote gives a face to an otherwise abstract business history. Software may help a publisher find readers. A founder still has to find someone willing to listen, then do enough useful work for that relationship to last. Trust cannot be added to a product with a conveniently labeled settings switch.
When making sentences became easier
By March 2015, Keywee had announced $9.1 million in funding, with Innovation Endeavors, Marker and The New York Times among its backers. The publishing business gave Makover something beyond a list of recognizable customers. It offered experience with the relationship between text, audience and response. That relationship would become central to Anyword.
Anyword launched in 2021. Later that year, it announced $21 million in new funding led by Innovation Endeavors. Its proposition joined AI-generated copy with predictions about performance. The timing matters. This was before ChatGPT made conversational AI a routine topic far beyond engineering teams. Makover was already trying to turn language-model capabilities into a working marketing product.
The early Toyota experiment belongs to that period of possibility. A model could produce something startling without producing something acceptable. A founder had to decide whether the result represented a dead end or a starting point. Makover chose to keep working on the distance between a machine’s response and a marketer’s brief.
Keywee’s founding year in Makover’s account
Anyword launches for marketers
New funding announced in November 2021
One thousand drafts, one decision
For Makover, the abundance of generated copy introduces a selection problem. Different audiences respond to different language; different campaigns ask them to do different things. An email open, a click and a purchase are distinct outcomes. The most elegant sentence in a meeting may have little to say about any of them.
In his account of Anyword’s workflow, a brief supplies the intended goal and context. Language models generate variations. A separate predictive model evaluates them using marketing performance data. The platform can also use a customer’s campaign history to build custom models. The point is to give a marketer something more specific to consider than an attractive turn of phrase.
A score remains a prediction. It needs an audience and a goal to mean anything useful, and the actual campaign still meets actual people. The distinction is practical. If a tool helps narrow the field, a team can spend more attention on the choices that remain. Nobody earns a customer’s interest merely by winning an internal argument about adjectives.
Anyword’s April 2022 announcement about analyzing $250 million in advertising reflected this interest in measurable language. Makover described improvements arising even from simple changes. The figure referred to advertising analyzed, rather than company revenue or money raised. It is evidence of the kind of material his business wanted to study: words attached to outcomes.
A voice with something to remember
Audience fit is only part of a marketer’s task. A company also wants its writing to sound like its own writing. In 2024, Makover publicly announced Custom Blog Models, allowing customers to upload previous blog posts and teach Anyword’s Blog Wizard their style. The useful input was concrete: examples of what the brand had already said.
This is an appealingly ordinary requirement for a technology associated with extraordinary claims. To learn a voice, look at the writing. A prompt that asks for warmth, clarity or confidence leaves considerable room for interpretation. Examples give those adjectives something to stand on. They also make the customer’s accumulated work part of the next draft.
Makover’s announcement ended with an admission that the post itself had been written with Anyword. There is a small joke in selling a writing tool through writing produced by the tool. There is also a straightforward demonstration. The product enters the founder’s own public communication, where people can read the result rather than merely hear the promise.
The work after the demo
By November 2025, Makover’s public discussions included enterprise adoption, marketing workflows and the consequences of effectively unlimited content. On SaaS Scaling Secrets, the conversation covered testing, data integration and the need to demonstrate returns while investing in longer-term AI infrastructure. Once the output becomes plentiful, the organization around it starts to matter more.
His January 2026 conversation with Avigad Oron and host Katrin Ribant carried that discussion into context management and enterprise architecture. A model sits inside a larger system. Relevant information must reach it; different components must work together; professionals need to understand enough of the process to judge its results. The demonstration is one part of the job.
These are less theatrical questions than whether a machine can write a paragraph. They are closer to a working day. Which information belongs in the brief? What should the team review? What does success mean for this campaign? A useful answer makes the next task easier. An impressive answer may simply create another task.
“Stand in luck’s way.”
A favorite saying Makover uses to describe preparation
The saying leaves room for circumstances without making preparation optional. Makover’s story contains years spent with language before it became an everyday technology conversation: studying it, building with it, and applying it to the way publishers and marketers reach people. His early machine needed better manners. The more capable machines need better direction.
The persistent question is small enough to fit inside a single campaign and large enough to sustain a company. Of all the words available, which belong here? Makover has spent much of his career approaching that question through software. The reader at the other end remains free to answer by paying attention, clicking, buying - or carrying on with the day.