--- title: AI Marketing Assistant on Approved Facts | Torke url: https://torke.tech/engine/context description: Torke Context gives your AI marketing assistant your approved facts, word rules and examples for every request, then checks each draft in code. --- # An AI marketing assistant on your facts Torke Context is a Torke Node, one part of Torke Engine, and you can buy or rent it on its own. It puts your approved facts, word rules and examples in front of your AI assistant for every request, loads them in code, and checks each draft against the same files before anyone on your team reads it. Diagram: How Torke Context works Torke Context runs in steps: your documents become cited claims, a named reviewer approves each claim, your brand is held as small files, code loads what the task needs. ## What is context engineering for marketing? Context engineering is the discipline that turns generative AI from an unpredictable collaborator into a creative partner whose inputs you control. For a marketing team it means deciding which facts, rules, examples and limits a model reads before it drafts an ad, an email or a page for your brand, loading them in code rather than by instruction, and checking the draft against the same material before anyone publishes it. ## Why does an AI marketing assistant guess? A model writes about your company from what the request gives it and what it absorbed in training. For a well-known company that training is broad and out of date, and for a smaller one it is close to empty, so when the right fact, rule or example is missing the model fills the gap with something plausible. Ask it for an ad with a strong proof point and it will produce a number, because a number is what a strong proof point looks like. We tested this on a worked example built only from a listed software company's public filings. A simulated first draft rounded the customer count to "300,000+" and placed the company in "140 countries", while the sources said 306,446 customers at 30 June 2026 and more than 135 countries. Both errors read as true, and only a check against the evidence file caught them. The same cause produces the four corrections marketing teams make again and again: invented figures, a voice that belongs to nobody, product names and prices that changed since the last brand book, and rules about claims, channels and approvals that only exist in someone's head. Pasting the brand book into the prompt does not fix it. In the same example the full brand context held about 36,000 tokens plus about 22,000 for the shared task rules, so sending everything costs roughly 58,000 tokens per request, while the pack built for one LinkedIn ad came to about 10,100 tokens, around 83 percent less, and still carried two approved example ads. Torke Context moves those corrections upstream, into files your assistant is given and checks every draft must pass. Your team stops correcting the same number twice, and when legal asks where a sentence came from, the answer is a file, a version and a date. ## How does Torke Context work? Each step says who does it. Anything with a written rule runs in code, a model works only where judgment on free text is needed and always under a code check, and a person decides whatever someone must answer for. ### Your documents become cited claims Who does it Code and model What happens Torke Intel reads your documents inside a sealed container with no network and writes each claim with the passage it came from, using a local model. Code rejects any claim whose quote cannot be found word for word in its source. ### A named reviewer approves each claim Who does it Person What happens In Evidence Review, a person on your team approves each claim with its audience, deal phase and expiry date. Every statement keeps a status, and only approved and reference content may support a claim in copy that leaves your company. ### Your brand is held as small files Who does it Code What happens Voice, word rules and product names, the catalogue of products and prices, and the evidence register each live in one file, so a fact exists in exactly one place and a corrected price cannot survive elsewhere in its old form. ### Code loads what the task needs Who does it Code What happens Each request is classified and code selects only the files that task needs, down to one product and its claims, then logs every file it loaded with its path, version and SHA-256 hash. ### Code checks the draft before a person sees it Who does it Code and model What happens Character limits, banned words, name spellings, claim status, numbers and colours are checked against the same files. A failed draft gets one revision with the report attached, and an AI review scores the one that survives. ### A person approves what leaves Who does it Person What happens Nothing is published, sent or spent by the system. A named person on your team approves every draft that leaves the company. ## What do you receive? ### Drafts that state only what you approved A number, a product name or a claim reaches a draft only if it exists in your files with a status that allows it. In the worked example, the claim check refused an ad built on a price the pricing page showed two ways. ### A record of what the assistant read For every draft you can see which version of each fact the model was given, which is the first question a legal team asks. ### The confirmations still missing Each draft lists what stands between it and publication. In the worked example that list held three items, and approving them was the whole remaining work. ### Questions ranked by what they unblock Open questions are ordered by how many tasks each one moves forward, so your first hour goes on the claims policy and the word list rather than on personas. ## What changes in your role? If you are the only marketer at a technical company, or the head of marketing whose numbers get questioned in every pipeline review, much of your week goes on answering for work you did not fully control: why a draft quoted the wrong figure, where a claim came from, whether legal ever saw it. Torke Context moves you to the other side of those questions. You stop checking what the assistant wrote line by line and start directing it, because you decide which facts, words and examples the brand stands on, and the checks run before a draft reaches you. When the founder, the CFO or a sceptical engineer asks where a sentence came from, you answer with the file, the version and the date the model read. You walk into the review as the marketing director who set the rules the drafts follow, rather than the person asked to defend each draft after the fact. We have not measured this on a client account yet, so read it as what the system is built to change. ## Where does it stop? Torke Context removes the errors caused by missing or unlabelled information. It does not make a weak offer persuasive, and it does not replace someone who knows your customer. In the worked example a draft that passed every code check still scored 3 out of 5 on audience relevance, because the persona behind it was built from public filings, and only interviews and CRM data fix that. Until someone on your side approves your facts, drafts can be reviewed but not published. That keeps every assumption visible in the files, the pack and the review, instead of buried in a live campaign. ## Who is it for? Who gets the most from Torke Context, and when another tool or no tool is the better answer. ### A good fit - Companies whose buyers check every claim, such as AI, robotics and data vendors, where one unsupported number can end an evaluation. - Teams whose claims carry regulatory weight, such as financial services, health products and real estate funds. - One marketing team running several markets and languages from the same facts. ### A poor fit - A company where nobody can approve facts, because the files then fill with assumptions that never become approved claims. ## Part of Torke Engine, and available on its own Torke Context is where every other node gets its facts. [Torke Reachout](https://torke.tech/engine/reachout) describes your company only through claims approved here, and the routing plan in [Torke Router](https://torke.tech/engine/router) sends its claim checks to Evidence Review. Approved claims also feed the campaign drafts your team runs from [Torke Engine](https://torke.tech/engine). Torke Context is a Torke Node, so you can run it on its own or inside Torke Engine with the other nodes, and add the others later without starting over. ## How much does Torke Context cost? Three ways to run it. Prices are in Swiss francs, excluding 8.1 percent VAT, and media or third-party data you choose to add is billed at cost. Run by Torke CHF 3’900 a month - We operate Torke Context for you inside Torke Service: the setup, every run, the checks and a named operator who answers for the output each week. [How Torke Service works](https://torke.tech/service) Rent it, where most teams start CHF 1’450 a month - Your team runs it from its AI assistant or the command line. Setup, the first supervised run, updates and support are included, and you can stop with 30 days' notice. [Rent Torke Context](#get-in-touch) Own it CHF 14’500 once - A licence to run it on your own infrastructure with no monthly fee, the setup session included. Updates and support are optional at CHF 2’900 a year. [Ask about owning it](#get-in-touch) ## Where does it stand today? Torke Intel and Evidence Review are built and run today. The code that builds a pack for each request and checks the draft runs from the command line with 24 passing tests on two complete example brands, and it is being moved into Torke Engine so your own assistant can request a pack directly. Onboarding starts from nine answers and one 45-minute validation session. ## What do teams ask about Torke Context? Each answer stands on its own, so you can forward one to a colleague, and an assistant asked the same question can quote it as it is. ### What is an AI marketing assistant? An AI marketing assistant is a language model your team uses to draft and check marketing work, such as ads, emails, landing pages and campaign plans. Its accuracy depends on what it reads with each request, which is why Torke Context supplies your approved facts, rules and examples and checks the result in code. ### Can we start without a brand book? Yes. You answer nine questions, including your name, website, colours, three words for your tone and the person who can approve facts. We read your public material, label everything we infer as an assumption, and a 45-minute session turns the assumptions you confirm into approved facts. ### Can it publish or spend on its own? No. It drafts, checks and proposes, and every ad, email or page waits for a named person on your team to approve it. ## The other Torke Nodes [ Torke Signal - ](https://torke.tech/engine/signal) [ Torke Visitor Signal - ](https://torke.tech/engine/visitor-signal) [ Torke Router - ](https://torke.tech/engine/router) [ Torke Reachout - ](https://torke.tech/engine/reachout) [ Torke Qualification - ](https://torke.tech/engine/qualification) [ Torke Ledger - ](https://torke.tech/engine/ledger) [Torke Engine - ](https://torke.tech/engine) ## Read next The article behind this node and the pages your colleagues will ask about. ### What marketing context engineering is Marketing context engineering decides what an AI model reads before it writes for your brand, loads it in code and checks the draft against the same material. How it works, what it costs and where it stops. [Read the article, 7 minutes](https://torke.tech/blog/marketing-context-engineering) ### Torke Engine Your own team runs the same engines from Claude, Codex or any MCP assistant. [How Torke Engine works](https://torke.tech/engine) ### Commitments Approved claims, sources, a named approver, a lawful basis and honest reporting. [Read the seven commitments](https://torke.tech/commitments)