{"id":1033,"date":"2026-05-14T13:55:21","date_gmt":"2026-05-14T10:55:21","guid":{"rendered":"https:\/\/blog.selector.space\/en\/?p=1033"},"modified":"2026-05-14T13:58:41","modified_gmt":"2026-05-14T10:58:41","slug":"artificial-intelligence-marketing","status":"publish","type":"post","link":"https:\/\/blog.selector.space\/en\/artificial-intelligence-marketing\/","title":{"rendered":"Artificial Intelligence for business: How to Avoid Overpaying for an AI Sticker"},"content":{"rendered":"\n

Artificial intelligence for business costs from $50 to $50,000 per month. Under the term \u201cAI\u201d 5 different technologies are sold: from real LLMs to ordinary if-else logic with a sticker \u201cAI-powered\u201d. In order not to overpay, you need to ask the vendor 7 specific questions about the technology.<\/p>\n\n\n\n

When a business owner says \u201cwe need artificial intelligence for business\u201d, he can mean completely different things: from a large language model (LLM) for thousands of dollars per month to simple conditional logic that a programmer can write in an hour.<\/p>\n\n\n\n

The problem is that the word \u201cartificial intelligence\u201d has long ceased to mean anything specific. It is a marketing name. It has been that way from the very beginning.<\/p>\n\n\n\n

How did the term “artificial intelligence” come about?<\/h2>\n\n\n\n

Who coined the term “artificial intelligence”?<\/h3>\n\n\n\n

The term was coined by John McCarthy in August 1955, not to describe a technology, but to apply for a grant.<\/p>\n\n\n\n

McCarthy was writing a proposal to the Rockefeller Foundation to fund a summer workshop at Dartmouth College. He was joined by Marvin Minsky, Nathaniel Rochester, and Claude Shannon.<\/p>\n\n\n\n

They needed a name that would accomplish three things at once:<\/p>\n\n\n\n

    \n
  1. It would sound ambitious for the foundation to allocate money for a “visionary” project<\/li>\n\n\n\n
  2. Differentiates from competitors: from Norbert Wiener’s cybernetics<\/li>\n\n\n\n
  3. Will be recognizable so that the scientific program crystallizes around it<\/li>\n<\/ol>\n\n\n\n

    McCarthy coined “artificial intelligence.” The foundation gave a grant. The workshop was held in the summer of 1956. The term stuck.<\/p>\n\n\n\n

    Did McCarthy really coin this term?<\/h3>\n\n\n\n

    No, not a fact. In a 1974 interview with historian Pamela McCordack, McCarthy admitted: he is not one hundred percent sure that he invented the term himself. Perhaps he heard something similar earlier from Wiener.<\/p>\n\n\n\n

    Even with the authorship of the term, everything is not as clean as they say at conferences.<\/p>\n\n\n\n

    What happened before the term \u201cAI\u201d?<\/h3>\n\n\n\n

    Before the Dartmouth workshop, the field had a working title: automata studies. That was the name of the collection that McCarthy and Shannon were preparing for publication in the same years.<\/p>\n\n\n\n

    \u201cAutomata studies\u201d sounded too technical. Foundations were reluctant to give money for it. And \u201cintelligence\u201d sounded completely different. A big word with a hint of the future. It sold better.<\/p>\n\n\n\n

    Conclusion: the word \u201cAI\u201d was born not in a laboratory, but in a grant application. It is an advertising slogan, not a description of the technology.<\/p>\n\n\n\n

    What will be sold under the name “artificial intelligence” in 2026<\/h2>\n\n\n\n

    How many different technologies are hidden behind the word “artificial intelligence for business”?<\/h3>\n\n\n\n

    At least five completely different things:<\/p>\n\n\n\n

      \n
    1. Real LLMs (large language models). Integration with APIs OpenAI, Anthropic, Google or self-hosted open-source models. Cost: from $50 to $50,000 per month depending on the volume.<\/li>\n\n\n\n
    2. Classic machine learning. Algorithms from the 1990s: recommender systems, classifiers, predictive models. Formally also AI, but nothing new.<\/li>\n\n\n\n
    3. Heuristics and expert systems. Rules written by a person that imitate “smart” behavior. Technology from the 1970s. Works, but no magic.<\/li>\n\n\n\n
    4. Ordinary if-else. Conditional logic with a sticker “smart” or “AI-powered”. Zero intelligence. Just programming.<\/li>\n\n\n\n
    5. Pure marketing. The product has not changed, but they added “AI” to the name, added neurons to the landing page in the illustration, and raised the price.<\/li>\n<\/ol>\n\n\n\n

      Why is this a problem for business?<\/h3>\n\n\n\n

      Under one word, technologies coexist that are as different as a moped and a truck. Both are transport, but prices and capabilities differ by orders of magnitude.<\/p>\n\n\n\n

      When the average customer hears \u201cservice with AI,\u201d he imagines something similar to ChatGPT<\/a>. An intelligent system that understands language and makes decisions. Technically, this is called LLM (Large Language Models).<\/p>\n\n\n\n

      But \u201cAI\u201d in advertising can mean something completely different. And without specifics, it is impossible to understand what you are paying for.<\/p>\n\n\n\n

      How to distinguish real AI from marketing?<\/h2>\n\n\n\n

      Comparison table: real AI vs marketing AI<\/h3>\n\n\n\n
      \"Comparing<\/figure>\n\n\n\n
      Criterion<\/td>Real AI (LLM\/ML)<\/td>Marketing-AI<\/td><\/tr>
      What’s under the hood?<\/td>Neural networks, transformers, trained models with billions of parameters<\/td>if-else rules, SQL queries, templates<\/td><\/tr>
      Adaptation<\/td>Learns from new data<\/td>Works only according to preset scenarios<\/td><\/tr>
      Documentation<\/td>Specific model (GPT-4, Claude), accuracy metrics<\/td>“Smart algorithm”, “AI-driven” without specifics<\/td><\/tr>
      Result<\/td>20-80% increase in metrics<\/td>0-5% or not measurable<\/td><\/tr>
      Cost<\/td>$3-5k (API) to $100+k (custom model)<\/td>Like regular software<\/td><\/tr>
      Subscription<\/td>Depends on volume (tokens, requests)<\/td>Fixed, overpriced “because AI”<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

      Real examples from the Ukrainian market<\/h3>\n\n\n\n

      Example 1: \u201cAI copywriter\u201d for $200\/month<\/h4>\n\n\n\n

      The Ukrainian service promised “text generation with artificial intelligence.” After checking, it turned out: a wrapper over the OpenAI API with a 300% markup.<\/p>\n\n\n\n

      The client could get the same for $20 (ChatGPT Plus) or $5-15 (direct API). Savings of $180\/month.<\/p>\n\n\n\n

      Example 2: \u201cSmart CRM with AI scoring<\/a>\u201d<\/h4>\n\n\n\n

      \u201cAI scoring\u201d turned out to be a table of 6 parameters, each giving 1-10 points. Sum = \u201cAI score\u201d. This is not AI. This is an Excel formula.<\/p>\n\n\n\n

      Example 3: Real AI that paid off<\/h4>\n\n\n\n

      The client asked for \u201csomething simple for FAQ, without AI.\u201d We checked the task: 70% of requests are non-standard. We deployed an LLM bot on Claude with RAG. Cost: $150\/month. We saved 20 operator hours every week.<\/p>\n\n\n\n

      Example 4: “AI Recommender” = sorting by category<\/h4>\n\n\n\n

      WooCommerce plugin for $80\/month showed \u201csimilar products\u201d by category and price. Algorithm from 2005. No ML.<\/p>\n\n\n\n

      Example 5: SEO texts \u201cwith AI\u201d without results<\/h4>\n\n\n\n

      SaaS generated texts via OpenAI API with the template \u201cwrite about [keyword]\u201d. Without strategy, semantics, competitive analysis. The texts were not ranked. The client paid $120\/month for a wrapper worth $8-10.<\/p>\n\n\n\n

      Regularity: the more the word \u201cAI\u201d and the less specifics, the more likely you are to pay for a sticker.<\/p>\n\n\n\n

      7 questions before buying artificial intelligence for business<\/h2>\n\n\n\n

      Ask these questions to any vendor selling an \u201cAI service\u201d:<\/p>\n\n\n\n

      If the vendor gets offended by these questions, it\u2019s a red flag.<\/p>\n\n\n\n

      \n

      1. What exactly is the technology under the hood?<\/h3>\n\n\n\n

      Need specifics: \u201cLLM from OpenAI,\u201d \u201cML model on scikit-learn,\u201d \u201crule-based engine.\u201d Not \u201csmart algorithm.\u201d<\/p>\n\n\n\n

      2. What are the system quality metrics?<\/h3>\n\n\n\n

      Accuracy, precision, recall, BLEU, user satisfaction score. If there are no metrics, the system is not measured, it does not work predictably.<\/p>\n\n\n\n

      3. How much better than the alternative without AI?<\/h3>\n\n\n\n

      Comparison with baseline. If the difference is 3%, there is no point in paying for \u201cAI\u201d.<\/p>\n\n\n\n

      4. How does it behave in non-standard scenarios?<\/h3>\n\n\n\n

      Real AI will give a reasonable attempt. A rule-based system will fail or produce a pattern.<\/p>\n\n\n\n

      5. Does it use our data for further training?<\/h3>\n\n\n\n

      Critical for privacy. Especially for Ukrainian jurisdiction with potential GDPR conflict.<\/p>\n\n\n\n

      6. How much does a provider’s API cost?<\/h3>\n\n\n\n

      If a vendor charges $200 and the OpenAI API costs $15 for your volume, you know the markup. Sometimes it’s justified (integration, support), sometimes it’s not.<\/p>\n\n\n\n

      7. Can we see a demo on our data?<\/h3>\n\n\n\n

      Real AI solutions are easy to demonstrate. Marketing AI is demonstrated with cherry-picked examples.<\/p>\n<\/div>

      \"Is<\/figure><\/div>\n\n\n\n

      Why don’t AI detectors (Copyleaks) work?<\/h2>\n\n\n\n

      Can Copyleaks and other AI detectors be trusted?<\/h3>\n\n\n\n

      No. The detectors are very unreliable.<\/p>\n\n\n\n

      OpenAI shut down its own AI detector in 2023<\/a> due to low accuracy. Stanford and University of Maryland research<\/a> shows:<\/p>\n\n\n\n