How to Measure AI Search Visibility Without Spending $500/mo

Why do most AEO measurement tools cost too much for what they deliver?

AEO tooling is in its early-2018-SEO-software phase. Platforms like AthenaHQ, Profound, and Goodie AI charge $499-$4,000/month for what is fundamentally a prompt scheduler, a response logger, and a brand-mention extractor. The math behind the price is API costs (small), engineering overhead (real), and willingness-to-pay from VC-funded mid-market customers (high). For a business under $200K MRR, the data-to-cost ratio is poor — you pay for a polished dashboard rather than for unique insight.

The actionable insight from any AEO tool is the same: which prompts mention me, which brands compete in those prompts, what sources are being cited. You can replicate that for the cost of API calls if you are willing to write 200 lines of Python and maintain a Google Sheet.

Tracking citation is worth doing. Seer Interactive analyzed 5.47 million queries across 53 brands and found that being cited inside an AI Overview delivers 120% more organic clicks per impression than appearing on the same SERP uncited — though still 38% fewer clicks than a SERP with no AI Overview at all (Seer Interactive, 2026). Seer is blunt about the limit of that finding: 'We cannot claim causation. Higher-authority brands are also more likely to be cited.' So citation is a number worth watching. It is not a number worth $500 a month to watch.

What does the manual baseline look like before any tooling?

The manual baseline is 30-50 prompts per week, run by hand across ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews, with results logged in a single sheet. It takes about 90 minutes per week for one person and produces cleaner data than most paid tools because you can read the full response and catch nuances (sentiment, framing, source attribution) that automated tools miss.

• Prompt list: 40-60 buyer-intent prompts across the 5 categories (BEST-X, comparison, gap/objection, problem, branded).
• Engines: ChatGPT (paid, for web search access), Claude, Gemini, Perplexity, plus a manual check on Google AI Overviews.
• Log per row: prompt, engine, date, your brand mentioned (Y/N), position if mentioned, competing brands named, sources cited, sentiment.
• Sheet: one tab per engine, one row per (prompt, engine, date). Pivot weekly for citation share trend.

What does Search Console tell you about AI visibility?

Less than you have been told. Google's own documentation states that sites appearing in AI features 'are included in the overall search traffic in Search Console' and are 'reported on in the Performance report, within the Web search type' (Google Search Central). There is no AI Overview filter. A click from inside an Overview lands in the same bucket as a click from a blue link, and you cannot pull them apart.

That kills a shortcut a lot of agencies sell. If a vendor shows you a dashboard of your 'AI Overview clicks' sourced from Search Console, they are modelling that number, not measuring it — ask them which field it comes from and watch what happens. What Search Console does give you is real and free: query-level impressions and clicks on the pages you built for AI extraction. Watch for impressions climbing on question-shaped queries while clicks stay flat. That gap is the zero-click fingerprint, and it is the closest honest proxy Google hands you.

It also explains why your rank tracker is not a substitute. Ahrefs analyzed 863,000 SERPs and 4 million AI Overview URLs and found that just 38% of pages cited in AI Overviews also rank in Google's top 10 for the same query — down from about 76% a year earlier (Ahrefs, March 2026). You can hold position 3 while the Overview sitting above you cites three sources you have never heard of. Ranking and citation have come apart, which is the whole reason this measurement problem exists. If you want that gap closed as a service rather than a spreadsheet, that is what our GEO program does.

How do server logs reveal which AI crawlers are visiting?

Every major AI engine crawls your site with an identifiable user agent. Server log analysis tells you crawl frequency, which pages are being visited, and whether your robots.txt is accidentally blocking the bots you want. A simple weekly log pull, charted per bot, is enough to track the trend. Get the names right — most of the bot lists circulating in SEO posts are two years stale and still list retired agents. These are the current ones, per the vendors' own documentation (OpenAI, Anthropic).

Two traps. First, blocking a training crawler is not the same as blocking a retrieval crawler — block GPTBot and you stay in ChatGPT search; block OAI-SearchBot and you disappear from it. Plenty of sites have blocked the wrong one. Second, Google-Extended and Applebot-Extended are training controls, not separate crawlers you will see hammering your logs, so do not expect a clean visit count for them.

A simple grep-and-count script over your access logs gives you a weekly bot-visit dashboard. Rising visit frequency for GPTBot and ClaudeBot is a leading indicator that your content is being absorbed into the next training cycle.

How do paid AEO tools compare on price and value in 2026?

• Otterly.ai — $29-$129/mo entry-level. Tracks ChatGPT, Perplexity, AI Overviews. Best for solo founders who want a dashboard, not a CSV.
• Peec AI — $99-$299/mo. Multi-engine + sentiment. Limited custom prompt taxonomy depth.
• AthenaHQ — $1,500-$4,000/mo. Enterprise-grade, full prompt clustering and competitive analysis. Worth it above $200K MRR with a dedicated owner.
• Profound — $499/mo entry, enterprise at $2,500+. Strong competitive analysis. Good for mid-market B2B.
• Goodie AI — bundled with managed-service offering only.
• DIY (Python + APIs) — $15-$20/mo. Best learning value. Best signal-per-dollar under $200K MRR.

What does the DIY Python script actually do?

The DIY script loops through your prompt list, fires each prompt against the OpenAI, Anthropic, and Gemini APIs (plus Perplexity if you have an API key), extracts brand mentions from each response using regex or a small classifier prompt, and writes the result to a Google Sheet or CSV. Total runtime is 15-20 minutes per weekly cycle for 100 prompts × 4 engines. API costs run roughly $15-$20/month at typical buyer-intent prompt lengths.

• Inputs: prompts.csv (your 40-60 prompts), competitors.csv (your brand + 5-10 competitor names), config (API keys).
• Loop: for each prompt × each engine, call the chat completion API with web-search enabled where supported.
• Extract: regex-match your brand name + competitor names in each response. Use a follow-up LLM call to classify position and sentiment.
• Log: append a row to a Google Sheet (via gspread) or CSV with prompt, engine, date, mentions, position, sentiment, cited URLs.
• Schedule: cron job on a $5/mo VPS or a GitHub Actions workflow on a weekly cadence.
• Dashboard: simple pivot in Sheets, or Looker Studio on top of the same dataset.

What is the cost breakdown of running the DIY stack monthly?

Price these against the model you are actually calling on the day you build it. Model names and per-token prices in this category change every few months, and a cost table copied from an old tutorial is how you end up budgeting for a model that no longer exists.

What dashboards should you build on top of the raw data?

Three dashboards cover most of the operational value. The first is citation share over time — a weekly line chart of (prompts where you appear) / (total prompts), split by engine. The second is competitor citation share for the same prompt set, so you can see whether you are gaining ground or simply riding category-wide growth. The third is source attribution — which third-party platforms are driving your citations, ranked by frequency. The third dashboard is the one that informs next moves: if 60% of your citations come from Reddit, double down on Reddit; if 40% come from Clutch, prioritize the Clutch playbook.

Citation share is the right headline metric because AI answers are not winner-take-all. Pew Research Center tracked the real browsing behavior of 900 US adults across 68,879 Google searches and found that 88% of AI summaries cited three or more sources (Pew Research Center, March 2025 data). You are not fighting for one slot. You are fighting to be one of several cited sources — which is a far more winnable game, and a completely different measurement problem from rank tracking.

The same Pew study is the reason you should not judge this program on clicks alone: when an AI summary appeared, users clicked a traditional search result on just 8% of visits, versus 15% when no summary was present, and clicked a link inside the summary itself on only 1% of visits. Being cited is often brand impression rather than a session. Measure it as such, and hold the channel to booked calls at the bottom rather than clicks in the middle.

How do you compare DIY data to paid-tool data for sanity-checking?

Run a 2-week side-by-side: take 20 of your prompts, run them through your DIY script and through a free trial of Otterly or Peec AI. Compare the brand-mention extraction and the citation share row by row. Where the two disagree, the disagreement is almost always an edge case — an ambiguous mention, a sentiment call — and a human reading the raw response settles it faster than either system will. The exercise has a second payoff: it tells you whether the paid dashboard is actually surfacing anything your sheet is not, before you commit to a subscription.

Upgrade to a paid platform when one of three conditions is true: your prompt list exceeds 200, you have multiple stakeholders who need a dashboard they can open themselves, or the engineering hours cost more than the subscription. Until then, the sheet wins. The wider strategy this measurement feeds sits in the AEO complete guide, and our retainers are published on the pricing page — no call required to see a number. If you want us to run the baseline for you and show you where you are cited today, get my free audit.