# Does AI mention your company, and how to measure it > One question repeated monthly answers it. Four measurements from our own site and an audit of what actually decides whether a model cites you. - **Canonical URL:** https://appwave.dev/en/blog/does-ai-mention-your-company - **Language:** en - **Updated:** 2026-08-20 --- - **Published:** 2026-08-20 - **Author:** Antoni Łubisz - Co-Founder, Frontend Developer - **LinkedIn:** https://www.linkedin.com/in/antoni-lubisz/ - **Reading time (min):** 6 - **Topics:** Marketing, AI Implementation, Risk ![Biała sinusoida i stos trzech poziomych pasków, obok których czwarty, szary pasek leży poniżej i poza stosem.](https://cms.appwave.dev/uploads/czy_model_ai_wymienia_twoja_firme_274a76c821.webp) You can measure it with one question repeated monthly: in how many of fifty questions your customer actually asks does the model name your company. We run that measurement on ourselves. In August 2026, asked what an AI implementation costs, the model cited seven sources, most of them small sites without high domain authority. The difference from classic SEO is not in the tooling. It is in what counts. In search you count positions and clicks. In a model's answer there is no position: there is a list of one to three names, and either you are on it or you are not. The measure is binary, which makes it easier to verify than to sell. Below are four measurements we ran on our own site, and a method you can repeat without buying anything. Every number is ours, with a date. ## Models already leave a trace in Search Console In our Search Console, between 15 May and 14 August 2026, there are six queries shaped in a way no human types: a phrase in quotation marks followed by a list of excluded social platforms. That is what a model's built in search sends when it goes looking for material. Our site appeared on them at positions one through ten. The quoted phrases show what the model was after: dev, founder, employee experience, a phrase about using AI for tasks, deloitte together with audit, and house. This is a free measurement you already own. Open the queries report, look for rows containing site exclusions, and see what the model actually reaches on your domain. Ours was instructive: the model lands on pages about the team and about practice, not on service pages. ## Models cite small sites, not only strong domains On 15 August 2026 we asked a search enabled model a customer's question: what an AI implementation costs, with specific price ranges and cited sources. The answer was thorough. It cited seven sources and pulled a concrete range from each. The cited domains were mostly small sites, without the authority usually treated as an entry requirement. That is the most useful finding in the whole measurement. The channel is open at small company scale, which is what separates it from commercial keywords in classic search, where the link barrier genuinely closes the door. You enter a model's answer on the quality of the answer, not on a link budget. ## On one query, AI Overview cited only what already ranked On 14 August 2026 we pulled the search results for a query about implementing AI in a company. Position one is an AI Overview, ahead of any organic result. Seven domains are cited. All seven were also in the organic top ten. On that query, citation was not a separate channel. It was a layer on top of ranking: no position, no mention. This is an observation from a single query and we treat it as one. It looks like AI Overview and classic results are coupled more tightly in this category than the popular "two independent channels" story suggests. We are checking whether the same holds on queries with a different intent. ## Index first, model optimisation second The order of those two decides the outcome, and it is easy to reverse, because content work is visible and presence in the index is not. We checked this across our own site, page by page. The method is simple and free: ask the search engine for a literal, unique sentence from your page's opening paragraph, in quotation marks. A page in the index always surfaces on its own literal sentence. If it does not come back, you have your answer before you change anything. Search Console gives the second confirmation. A URL with zero impressions across three months, while neighbouring URLs in the same section collect them normally, has a presence problem rather than a content problem. On our site that test flagged one URL against sixteen working ones, so it was immediately clear this was a single page rather than the whole blog. The rule that follows is simple. A search enabled model draws its sources from the index, so a page outside the index will not be cited no matter how well it answers. **Before you optimise content for models, check whether that content exists for the search engine at all.** ## How to measure this yourself, without buying tools Write down fifty questions your customer actually asks. Not keywords. Questions in the form they arrive in on sales calls and in messages. We keep ten for each of five customer groups. Ask each one of a search enabled model and record one of three values. Two, if your company is named and described in a sentence that would not fit a competitor. One, if it is named with a generic description. Zero, if it is absent. Telling a one from a two carries the whole meaning of this measurement. You can be present and indistinguishable: the model lists you among providers and describes you in a sentence that fits every other name on the list, while the competitor next to you gets something specific. More ones at a steady number of twos means presence is growing without difference. Repeat the same list every month. A list you extend as you go measures its own changes rather than changes in visibility. ## What this measurement does not include There is no forecast. We cannot tell you how many citations will appear next quarter, and we do not know anyone who can calculate that honestly. We do not manipulate model answers. Search engine guidelines treat that as spam in the same sense as manipulating rankings, and beyond that it does not last. We also do not add structured data "for models", because controlled tests show models ignore it. What we have serves the search engine and stays. We do not sell an \`llms.txt\` file as part of a visibility plan. We have one and generate it automatically, but the search engine ignores it, so it is not an argument in a proposal. ## Before you write to us We do not promise rankings or mentions in models. We also do not take on projects where visibility in AI answers is meant to replace visibility in search, because on the query we checked, only pages already in the organic top ten were cited. If your site has an indexing problem, we start there and say so in the first meeting, even when the question was about something else. If you want to know your own starting point before changing anything: [see what an AI audit covers](https://appwave.dev/en/services/ai-audit). The first output is a number, not a recommendation. ## Questions & Answers (FAQ) ### How is optimising for models different from SEO? The metric. In SEO you count position and clicks. Here you count in how many of fifty questions your company is named. The tooling and the technical foundation largely overlap, because a search enabled model takes its sources from the search index. ### Can a small company get cited by a model? Yes, and we measured it: answering a question about implementation costs, the model cited seven sources, mostly small sites without high domain authority. The entry barrier is lower here than on commercial keywords in classic search. ### How do I check whether models reach my site at all? In the Search Console queries report, look for rows with a quoted phrase followed by a list of excluded social platforms. That is the shape of a query sent by a model's built in search. We had six of them across three months. ### How long before a company starts getting mentioned? We do not give a timeline and do not promise citations, because we control neither the index nor how a model picks sources. What we do give is a starting number, measured monthly against the same list, so after two measurements you see direction instead of impression. --- ## About AppWave AI automation agency from Łódź, Poland. We build custom AI software, automation agents and web applications. We co-create them with the client and guarantee every system we deliver. - **Legal name:** AppWave sp. z o.o. - **Address:** ul. Kolumny 147E/1, 93-611 Łódź, Poland - **Phone:** +48 538 441 413 - **Email:** office@appwave.dev - **Business hours:** Monday-Friday, 09:00-17:00 - [Free consultation](https://appwave.dev/en/appointment)