# Measure AI search ROI for a B2B SaaS site > Count the clicks, citations, and trials that AI search drives to your site so you can tell whether GEO is worth paying for before you ask for a quote. Source: https://comppound.com/blog/measure-ai-search-roi-for-a-b2b-saas-site (published 2026-10-16) What AI search costs you depends on three things you can count before asking for a quote: how many pages AI systems could cite, how often your buyers arrive through an AI answer, and how many of those visits turn into trials or demos. Count those inputs first and you can compare two proposals line by line instead of guessing which number is fair. Say you run a thirty-person SaaS company selling an API monitoring tool. You have a blog with a hundred posts, some landing pages, and no tracking for AI-driven traffic. A provider quotes you a monthly fee for GEO work. Before you sign, you need to know what you are paying for, what the return looks like, and whether the channel can produce it at all for your product. Start with the pages AI can cite. AI answers do not link to every page on a site. They cite pages that answer a specific question clearly, that have structured data where it helps, and that appear in sources the model trusts. Count how many pages on your site could realistically answer a question a buyer would ask an AI tool. That number sets the ceiling on how much citation volume is possible. A site with ten useful pages cannot produce the same citation volume as a site with a hundred, so a quote that assumes high volume from a small site is either padding the scope or overpromising. Next, count the visits that actually arrive from AI search. Most analytics tools do not label AI traffic as its own channel. It usually shows up as direct or referral with no campaign tag. Look for visits from domains you do not control that land on blog posts or comparison pages, especially from referrers you do not recognize. Set up a report that isolates those sessions so you can see the baseline before you pay to improve it. Without a baseline, any improvement a provider claims is impossible to verify. Then count what those visits do. A visit from an AI answer is only valuable if it leads to a trial signup, a demo request, or another step your sales team can work. Connect your analytics to your conversion events so you can see whether AI-referred visitors convert at the same rate as search-referred visitors. If they do not convert, paying for more AI visibility is paying for traffic that does not pay for itself. The cost direction follows from those three numbers. More citable pages means more production work, which pushes the price up. A higher baseline of AI traffic means less ground to cover, which can push it down. A low conversion rate means you need more volume to reach the same revenue, which pushes the required investment up before the channel becomes worth it. A fair proposal includes a defined scope, a measurement plan, and a review date. It should say exactly which pages will be optimized, how AI-referred traffic will be separated from other traffic, which conversions count, and when you will look at the data together to decide whether to continue. A bad proposal leaves out the measurement plan. It promises visibility or citations without saying how either one will be tracked, and it has no date at which either side can walk away if the numbers do not move. Ask a provider these questions before signing: - How will you separate AI-referred traffic from direct and referral traffic in my reports? - Which pages on my site are you targeting for AI citations, and why those pages? - What conversion events will you use to judge whether the traffic is valuable? - At what point will we review the data and decide whether to continue? It is not worth paying for AI search visibility if your site has almost no pages that answer a buyer question, if your analytics cannot isolate AI traffic, or if AI-referred visitors never convert. In those cases the money goes to production with no way to prove it came back. Fix the measurement and the page quality first, then pay for growth. [See how we do it](/services/seo-growth) ### Audit your pages for AI relevance 1. List every top‑level page that receives traffic from the blog, product, or docs sections. 2. For each page, ask whether a buyer could type a question that the page answers *in one sentence*. If the answer requires a multi‑step tutorial, the page is a poor AI citation candidate. 3. Check for structured data (FAQ, How‑to, or Product schema). Use the Google Rich Results Test; a page that validates schema is more likely to be selected by a language model. 4. Score the page on a simple 0‑2 scale: 0 = no answer, 1 = partial answer, 2 = complete answer with schema. Sum the scores; a total of 80‑100 points for a 30‑person SaaS suggests a healthy citation pool. If the score is low, focus first on expanding or rewriting a handful of high‑value pages before any provider can deliver results. ### Isolate AI‑driven traffic in your analytics AI referrals rarely appear as a separate channel. To extract them: - **Log file analysis** – Pull server logs for the past 30 days and filter for referrers that are not your domain and do not match known ad networks. Typical AI referrers include `*.search.googleapis.com`, `*.bing.com` with a `?q=` parameter that contains “how to” or “what is”. - **Custom dimensions** – In Google Analytics, create a dimension called “AI Source”. Flag a session when the landing page matches one of the pages you scored as a citation candidate *and* the referrer matches the patterns above. - **Conversion path comparison** – Compare the bounce rate, time‑on‑page, and downstream events for AI‑flagged sessions versus organic sessions. Even a small divergence (e.g., AI visitors spending 30 seconds longer) can justify the extra investment. Document the baseline number of AI sessions per month; this becomes the denominator for any ROI calculation you discuss with a provider. ### Benchmark conversion performance AI traffic tends to be more intent‑focused because the answer is often the last step before a purchase decision. To test whether that intent translates into value: - Set up two conversion goals: **Demo request** and **Free‑trial signup**. - Track the conversion rate for AI‑flagged sessions and for standard organic sessions over a 4‑week period. - If AI conversion is within 10 percentage points of organic, you have a viable channel. If it is dramatically lower, the content may not be aligning with buyer intent, and you should improve the page copy before scaling. Use these conversion rates to estimate the volume of AI traffic needed to hit a target number of qualified leads. For example, a 2 % conversion rate means you need 500 AI sessions to generate 10 demo requests. ### What a fair proposal should contain - **Scope matrix** – A table (in the proposal document, not here) listing each target page, the agreed‑upon citation goal (e.g., “appear in 3 AI answers per month”), and the production effort estimate. - **Measurement plan** – Exact definition of AI traffic, the analytics dimensions you will use, and the frequency of reporting (weekly dashboards, monthly deep‑dive). - **Milestones** – A timeline with checkpoints: 1) initial audit completion, 2) first citation delivered, 3) baseline traffic split, 4) conversion review. Each milestone should have a go/no‑go decision gate. - **Pricing breakdown** – Separate the *production* cost (content creation, schema markup) from the *optimization* cost (prompt engineering, citation outreach). This lets you see which line item grows as you add more pages. A proposal that mixes the two without explanation is a red flag; it makes it impossible to attribute results to the work you actually paid for. ### Additional questions to vet a provider - How do you decide which AI models (ChatGPT, Gemini, Claude) to target for citations? - What is your process for validating that a citation actually appears in an answer? - Can you share a sample report that shows AI‑traffic separation and conversion attribution? - How do you handle a scenario where a citation is removed by the model after an algorithm update? These questions surface the provider’s monitoring capability and their willingness to adjust tactics when the model’s behavior changes. ### When the investment is not justified 1. **Insufficient citation pool** – Fewer than 10 pages score as complete answers; the ceiling on volume is too low to achieve meaningful lead numbers. 2. **No baseline AI traffic** – If your log analysis finds fewer than 5 AI sessions per month, any claimed uplift cannot be measured. 3. **Conversion dead‑end** – AI visitors consistently bounce without triggering any downstream event after three months of tracking. In those situations, allocate resources to content creation and analytics setup first. Once you have at least 20 citable pages, 50 AI sessions per month, and a conversion rate comparable to organic traffic, then revisit a GEO provider. Comppound helps teams build the tracking and page scope needed to measure AI search visibility before they spend on production. [See how we do it](/services/seo-growth). ## Frequently asked questions ### How do I know if AI search is sending traffic to my site? Look in your analytics for sessions that arrive from domains you do not own and land on informational pages, especially referrers you do not recognize from traditional search. AI traffic often appears without a campaign tag, so it can hide inside direct or referral. Create a segment that isolates those sessions so you have a baseline before paying to improve it. ### Which pages should I optimize for AI citations? Prioritize pages that answer a specific question a buyer would ask, such as how a feature works, how two tools compare, or what a setup involves. AI systems cite pages that give a clear, direct answer rather than pages that only describe the company. Match the page to the question before worrying about formatting. ### What counts as a return from AI search? Count the visits that arrive through AI answers and then complete a conversion your sales team can use, such as a trial signup or demo request. If those visits do not convert, more visibility does not create revenue. Tie the traffic to your existing conversion events before judging the channel. ### When should I not pay for AI search optimization? Do not pay when your site has few pages that answer buyer questions, when you cannot separate AI traffic from other sources, or when AI-referred visitors do not convert. In each case you would be paying for production with no way to verify a return. Fix the measurement and content gaps first.