Free ways to check your brand's AI visibility

The case for checking at all is now a data point rather than a hunch. G2's 2026 buyer research found that 51 percent of B2B software buyers start their research with an AI chatbot more often than with Google, up from 36 percent just seven months earlier. Forrester's 2026 Buyers' Journey Survey of 18,000 buyers found 94 percent used AI during their most recent purchase, and 55 percent compared vendors inside AI tools before talking to anyone. If your category's buying questions are being answered inside a chat window, the brands named in those answers are on the shortlist before your website ever loads.

The stakes of being absent are equally concrete. The same G2 research found that 69 percent of B2B software buyers chose a different vendor than they originally expected because of AI chatbot output, and 33 percent bought from a brand they had never heard of before the AI named it. Meanwhile 51 percent of B2B tech brands have zero citations across ChatGPT, Perplexity and Gemini. Half the market is invisible in the channel where a majority of buyers now start. A free check tells you which half you are in, and that answer is worth an afternoon of anyone's time.

One framing note before the methods. AI visibility is a measurement problem first, and sloppy measurement is worse than none because it produces confident wrong conclusions. Each free method below comes with its specific failure mode spelled out, so you know exactly how far to trust what it tells you. For the deeper methodology, the companion piece on measuring AI visibility without lying to yourself covers the full method.

The most direct check costs nothing but time. Write down fifteen to twenty questions a buyer in your category would actually type: best tools for the job you do, alternatives to the incumbent everyone knows, how to solve the problem your product solves, and comparison questions between rivals. Comparing vendor strengths and weaknesses is the single most common B2B use of AI chat, at 41 percent according to G2, so comparison phrasings belong on the list. Then open ChatGPT, Perplexity and Gemini, ask each question in a fresh conversation, and record three things per answer: which brands were named, in what order, and which sources were cited.

Two rules keep the exercise honest. First, use fresh chats for every question, because earlier turns in a conversation steer later answers and you want the cold-start response a stranger would get. Second, log the full answer text, in a spreadsheet or a document, and never just a yes or no. The sources an engine cites for a question you lose are your action list later: those are the pages that would need to mention you for the answer to change. A full manual protocol, with a question template and a scoring sheet, is in our step-by-step AI visibility audit guide.

The limit of this method is repeatability. You will do this once, maybe twice. Nobody manually re-asks twenty questions across three engines every week, and as the next section explains, a single run is the least trustworthy sample you can take. Treat the manual check as a first look at the terrain, and treat any single surprising result, good or bad, as provisional until you have seen it repeat.

Here is the problem with asking once. SparkToro measured the consistency of repeated ChatGPT brand recommendations and found under a 1 percent chance that two identical runs return the same list of brands. The answers are probabilistic by design. Ask the same question five times and you will typically see a stable core of brands that appear in most runs, plus a rotating cast that appears in one or two. If your brand shows up in one run out of five, a single manual check has a good chance of telling you either that you are visible or that you are invisible, and both readings would be wrong.

The practical fix costs nothing except patience: ask each question three to five times, in fresh chats, and score presence as a fraction rather than a verdict. Appearing in four of five runs is real visibility. Appearing in one of five is a toehold. Zero out of five is a genuine gap. This turns a misleading coin flip into a rough frequency estimate, which is the honest unit of AI visibility. The full explanation of why answers vary this much, and what it means for measurement, is in why AI gives a different answer every time you ask.

This is also the honest pitch for automation, so we will make it plainly. Repeated runs with stored answers is exactly what a tracking tool is: Reachroller runs your question set on schedule through official engine APIs, stores every raw answer, and scores visibility as a trend line rather than a snapshot. The measurement method is the same one you would use by hand. The tool just makes rerunning it every week economically sane.

The second way free checks go wrong is question selection. If you ask an engine what is Acme and is Acme any good, the answer will mention Acme by construction. Every branded question you include inflates your apparent visibility, because the engine is answering about you rather than choosing you. The number that predicts revenue is unbranded visibility: how often you get named when the buyer does not mention you, because that is the moment a shortlist forms and the moment 33 percent of G2's surveyed buyers discovered the brand they eventually bought.

Branded questions still belong in the audit, in a separate bucket. They tell you whether the engines describe you accurately: right pricing, live features, no confusion with a similarly named company. Errors there are a different problem with a different fix. But keep them out of your headline score. This distinction is baked into Reachroller's scoring, which excludes branded questions from the headline number automatically, and it is explained in depth in branded vs unbranded prompts. When you evaluate paid tools later, ask each vendor how they handle this. Many do not document it, and an undocumented answer usually means the inflating kind.

If you want a number without the spreadsheet work, HubSpot's AEO Grader is the best-known free option. It is a one-time check with no account required: it queries ChatGPT, Perplexity and Gemini about your brand and returns a composite score out of 100 within a few minutes. As a zero-effort baseline it is genuinely useful, and it is a reasonable artifact to put in front of a skeptical boss, because a branded score of 34 out of 100 starts a budget conversation faster than a spreadsheet does.

Know what you are getting, though. It is a snapshot, and as the volatility section explained, any single-run reading of a probabilistic system carries wide error bars. It grades your brand as a whole rather than tracking the specific buying questions you win and lose, so it cannot tell you which answer to fix first. And the free tier is deliberately a top of funnel for HubSpot's paid monitoring, which runs $50 per month or comes bundled with Marketing Hub Pro. Run it, keep the score, and treat it as one data point beside your manual check rather than a verdict on its own.

The third route is using a paid tool's trial as a free structured audit. This gets you what the manual check and the one-time grader both lack: repeated runs, stored answers, and a question-by-question breakdown, at least for the trial window. Reachroller's trial is three days with 50 credits, every feature unlocked and no card required. One credit is one AI answer, so 50 credits covers a full first report across your question set, with enough left to generate one publish-ready fix page for a question you lose, since a generated fix costs ten credits. You leave the trial with a real baseline and a concrete artifact, whether or not you ever pay.

Other vendors offer trials too, and the honest comparison is short. Otterly.AI gives seven days across its seven monitored surfaces, and it is a capable monitoring product, named a 2025 Gartner Cool Vendor with more than 30,000 users. The difference is what you hold at the end: Otterly's trial ends with a report, Reachroller's ends with a report plus a generated fix you can publish. Note the vendors that offer no trial at all. Semrush's AI Visibility Toolkit has none, and Profound is demo-led with no public checkout. When a free audit is the goal, the trial-friendly end of the market is where to look, and the wider comparison in our full tool guide covers the whole field.

A tactical note: prepare your question list before you start any trial clock. Three days is plenty for a baseline if you arrive with twenty good unbranded questions ready to paste, and wasted if you spend day one deciding what to ask. The manual method above doubles as trial preparation, which is one more reason to do it first.

Your analytics account already holds free evidence of AI visibility: referral traffic. Sessions arriving from chatgpt.com, perplexity.ai, gemini.google.com and claude.ai mean an AI answer cited one of your pages and a human clicked through. In GA4, build a simple exploration filtered to those referral sources, or create a channel group for AI referrals so the trend is visible on one chart. The volume will look small at first. In some measurement panels AI referrals are still around 0.18 percent of all sessions, so do not expect a flood.

Small does not mean unimportant, on two grounds. First, growth: WebFX analyzed 2.3 billion sessions across 2024 and 2025 and found AI traffic grew 796 percent, and SE Ranking's study of 101,574 websites found ChatGPT referrals jumped 36.7 percent in May 2026 alone to an all-time high. ChatGPT commands roughly 92 percent of trackable LLM referral traffic, so if you watch one source, watch that one. Second, intent: WebFX found AI-referred visitors converted around 1.2 times organic, Semrush's 2026 analysis measured roughly 4.4 times standard organic, and Adobe found AI traffic converting 42 percent better. The studies disagree on magnitude and agree on direction: these visitors arrive pre-sold by an answer that already recommended you.

The limit is structural. Analytics counts clicks, and AI answers resolve many journeys with no click at all: the buyer reads the recommendation and types your name into Google later, landing in direct or branded search where the AI origin is invisible. So read referral data as a floor on your AI visibility, never a ceiling, and never conclude from quiet referral numbers that AI answers do not matter in your category.

One more free signal lives in your server logs or CDN dashboard: AI crawler visits. OpenAI operates OAI-SearchBot for its search features, and a Botify analysis found OpenAI roughly tripled its web crawl since August 2025. Perplexity, Anthropic and Google each run their own bots. Filter your logs for those user agents and you learn two useful things: whether the engines can reach your site at all, and which of your pages they pull most often. A robots.txt rule that blocks these crawlers, sometimes left over from an early decision about training data, quietly removes you from the search side of AI answers too.

Being crawled is necessary but never sufficient: it means your pages are available to be cited, and says nothing about whether any answer actually cites them. Treat crawler activity as a plumbing check. If OAI-SearchBot has never hit your site, fix access before anything else. If it visits regularly and you still lose every answer in your category, the problem is your content's citability, which is a different fix and the one covered in how to write content AI engines actually cite.

Run all five methods and you will know, with reasonable confidence, where you stand today. Three things remain out of reach. The first is the trend. AI visibility moves as engines refresh their retrieval, competitors publish, and models update, and a baseline from July says little about October. Trend lines require the same questions asked on a schedule with answers stored, which is sustained work no free method covers.

The second is attribution depth. A free check shows you lost a question. It rarely shows the pattern across fifty questions: which rival is winning your category's answers, from which cited sources, and whether the mentions come from the model's training data or from live retrieval, a distinction that decides whether you can change the answer in weeks or must wait for a retraining cycle. The third is the fix itself. Every method in this guide is diagnosis. Changing a lost answer means publishing a page an engine can cite, getting it indexed, and rechecking, and that loop is where free ends in every tool's pricing, including ours.

Our recommendation is the order this guide already follows. Spend the afternoon on the manual check, run the free grader for a second opinion, and read your analytics floor. If what you find is comfortable visibility, recheck quarterly and spend your budget elsewhere. If you find the gap most brands find, and 51 percent of B2B tech brands have zero AI citations, the cheapest serious next step is a trial built for exactly this: Reachroller's three days and 50 credits produce the tracked baseline and the first generated fix, no card, and $29 per month after that only if the loop earns it. Details are on the pricing page.