strawbi

Questions about AI assistants, your traffic and your site

Plain answers for small marketing, growth and devrel teams, including where strawbi helps and where another tool does the job.

Are AI assistants sending anyone to my website?

Your web analytics answers that, not strawbi. Look at referrals from the assistants' own domains, such as chatgpt.com, perplexity.ai, claude.ai and gemini.google.com, and compare them with Google over the same weeks. Treat the count as a floor: some assistants and apps don't pass a referrer, so those visits arrive looking like direct traffic.

strawbi shows the step before the visit: whether the answers to your buyers' questions cite your pages, and which ones. If you run devrel and your docs traffic looks off, that tells you whether your docs pages are the ones being cited, and whether coding agents can read them at all (robots.txt, text without JavaScript, markdown versions, llms.txt).

Our search traffic is falling. Where do I start?

Start by finding out whether something broke on your side or whether people are searching differently.

  1. Google Search Console: did impressions fall, or only clicks? Impressions holding while clicks fall is the pattern people describe when answers on the results page take the click.
  2. Your own pages: check what changed. strawbi's changes list shows every page whose text changed, pages that started redirecting or returning errors, and robots.txt edits, each with the date.
  3. The crawlers: make sure robots.txt or a bot wall isn't blocking the AI crawlers. strawbi checks 16 of them on every crawl.
  4. The answers: ask the questions your buyers ask and see whether the answers cite you or a competitor. strawbi does this every day.

strawbi doesn't show traffic itself. It covers steps 2 to 4; Search Console and your analytics cover the visits.

What do I tell sales when they ask if we show up in AI answers?

Give them a count, not a yes or no. strawbi's overview leads with one sentence you can paste, something like “cited in 282 of 671 answers”, over the last 14 days, against the 14 before. Under it are the questions where a competitor was cited and you weren't, which is usually what sales really wants to know.

Add one caveat: we ask through the assistants' search APIs, which is close to what a buyer sees in the app but not the same, because the app adds the buyer's history and location.

Leadership wants a number for how often AI assistants mention us.

You can give them two: how often the answers cite one of your pages, and how often they name you. strawbi reports each as a range with the answers behind it (“34–49%, cited in 70 of 168 answers”), because the same question gets different answers from one run to the next, and a single percentage would move on noise alone.

Two things to tell leadership up front: a figure appears once there are 30 answers on an engine, and strawbi calls something a rise or a fall only when this fortnight's range and the last one's don't overlap. For a devrel team showing how agents find the docs, add the readability report: which AI crawlers can get in, and whether the docs have text without JavaScript and markdown versions for coding agents.

Can I track brand mentions in AI answers?

Yes. For every answer, strawbi records whether your name (or an alias you add) appears and whether one of your pages is cited, and the same for each competitor you confirm. It suggests competitors from the sites the answers actually cite, with how often. Today the questions run on Claude and OpenAI's models; Perplexity and Gemini aren't switched on yet.

Nobody can switch it on for you, and anyone who promises to is guessing. What the evidence so far points to, from strongest to weakest:

  1. Be readable. If robots.txt blocks the AI crawlers or your pages need JavaScript to show text, an assistant that searches can't use them. strawbi checks this on every crawl.
  2. Get onto the pages that are already cited. Answers often cite lists, reviews and forum threads rather than vendors' own sites. strawbi lists the pages cited where you weren't, so you know whose list to get onto.
  3. Say plainly what you are and who it's for. Keep pricing and features current, put the answer near the top of the page, and compare yourself honestly with the alternatives.

What hasn't shown an effect in controlled tests: llms.txt and structured data. They're still useful for coding agents and assistants that browse live, but they won't get you recommended on their own.

Perplexity isn't one of the engines we ask yet, so strawbi can't show you its answers today; the steps above apply to it too. After a change, watch the answers for a couple of weeks before deciding it worked: strawbi marks your edit on the chart and calls a move only when the ranges stop overlapping.

ChatGPT tells developers wrong things about our product. How do we fix it?

You can't edit an assistant's answer, but you can fix what it reads. When an assistant searches, it reads your pages and other people's; when it answers from memory, it repeats what it learned in training, which only changes when the model is retrained.

  • Put the current facts on your own pages: pricing, features you've removed, the current version of the API.
  • Make the docs easy for agents to fetch: no robots.txt block, text without JavaScript, and markdown versions or llms.txt for coding agents.
  • Find the pages the wrong answer cited. An old blog post of yours can be updated; a third-party page means asking its author.

strawbi keeps every answer to the questions you track with the pages it cited, so you can see where the wrong fact came from and whether the answer changes after you fix it.

Does this matter for a B2B company under 50 people?

It matters once your buyers ask AI assistants for options before they search, which is when a shortlist of three tools gets written without you on it. Visits from assistants are still small: across 1,200 publisher sites, Microsoft Clarity found they were under 1% of traffic in late 2025. So this is about being on the shortlist rather than about volume. Small companies have one advantage: the fixes are small, often a headline, a robots.txt line or one page that says what you do.

What does it cost?

strawbi is free while we build it, and prices aren't set yet. We'll say what it costs before we charge anyone.

For comparison, when we read the AI visibility trackers' pricing pages on 21 October 2026, their entry plans ran from about $29 to $99 a month for 15 to 50 questions. See how strawbi compares.

We have no SEO hire. Is it worth starting yet?

Start if one of these is true: buyers have told you they found a competitor through an AI assistant, someone above you is asking whether you show up, or search traffic is falling and you can't say why. If none is, it can wait.

Starting early has one advantage: strawbi's record of your pages and answers begins the day you add the site, and it can't be filled in backwards. Today it costs the time it takes to add your site and read the report.

Add your site to see the first report.