Skip to main content
AI-visibility

Measuring AIO and GEO: Why Prompting Agencies Fail and How Deterministic Engineering Closes the Analytics Blind Spot

Anthony Skinner, CEO and Co-Founder, and the Pressfit.ai Engineering Team12 min read

The Agency Trap: Prompters vs. Engineers

Most marketing agencies and consultants selling AI optimization (AIO) and generative engine optimization (GEO) are failing their clients, because they treat a data engineering problem like a creative writing exercise.

Right now, the standard agency playbook relies on prompters: strategists typing queries into ChatGPT or Perplexity by hand, guessing why a competitor showed up, and rewriting a few paragraphs of copy. There is no magic bullet or secret prompt that forces an LLM or Google's AI Overviews to recommend your brand. LLMs are probabilistic, so the same prompt can return different citations from run to run, and manual spot-checking produces false positives and wasted retainers.

At Pressfit, we reject the magic-bullet approach. Showing up in AI search means connecting the dots across large, structured datasets:

  • Your website's architecture, technical health and existing content.
  • Your competitors' domains, authority signals and content gaps.
  • The search terms and conversational intents that drive buyers to your site.

Real GEO takes time, and most agencies fail to deliver it, because automating what a multi-person agency team would spend weeks researching by hand takes deterministic compute. Instead of guessing with prompts, you build repeatable pipelines that ingest site structures, map entity relationships, cross-reference search intent and output a prioritized playbook. Most agencies don't have data engineers. They have prompters. Without deterministic engineering under the hood, they can't execute at the scale or consistency AI engines require.

You can't prompt your way into an LLM's retrieval. In March 2026, Ahrefs found that only 37.9% of URLs cited in Google AI Overviews also ranked in the top 10 results, and 31% didn't rank in the top 100 at all, down from about 76% from the top 10 in mid-2025 (Ahrefs). AI engines pick sources for entity clarity, structured topical coverage and how well a page answers the follow-up questions behind a query, not for keyword stuffing.

WorkflowTraditional agency (prompters)Pressfit Playbook (deterministic compute)
Competitor and site auditClicking through a few competitor blogs by hand and guessing why they rank.Crawling and parsing the page structure and schema of your site and your competitors' sites to map gaps at scale.
Intent and query mappingSpot-checking a handful of broad prompts in a single chat window.Working from real search demand, mapping the follow-up questions behind each query and clustering commercial buyer intent.
Entity and authority signalsRewriting copy with "AI-friendly" buzzwords and hoping a model picks it up.Extracting named entities, co-occurrence patterns and third-party brand mentions, which Ahrefs found correlate about three times more strongly with AI Overview visibility than backlinks.
Execution and outputSubjective advice ("write more authoritative content") that varies by strategist.A deterministic, prioritized playbook with repeatable scoring rules, so the deliverable is consistent and auditable every time.
Why it scales (or fails)Breaks under client load; human spot-checks can't keep pace with model updates.Takes engineering time up front, then delivers predictable, compounding authority across Google AI Overviews and LLMs.

The Measurement Problem: Why Everyone Claims "It Can't Be Tracked"

Once you do the engineering work to show up in AI Overviews and LLMs, you hit the second wall: attribution.

A vocal part of the SEO industry says AIO and GEO can't be measured, and uses that as a reason to avoid accountability. Traditional rank tracking is broken, but throwing up our hands isn't an option. Buyers search differently than they used to, and measurement has to adapt.

The 2026 data shows both sides. Trackable AI referral volume looks small because of attribution blind spots, yet the traffic that does come through is changing pipeline quality.

  • Zero-click is the default. From January to April 2026, 68% of U.S. Google searches ended without a click, according to Similarweb data analyzed by SparkToro. That panel excludes the Google mobile app, so the real share is likely higher (SparkToro).
  • Citations win the clicks that remain. In Datos clickstream data, the first source cited in an AI Overview takes 47% of the clicks that leave the panel, the second 23% and the third 14%, so the top three take 84% (Mintec). Seer Interactive found brands cited in an AI Overview averaged a 2.07% organic click-through rate on informational queries, against 0.94% for brands that weren't cited.
  • The referrer blind spot. Across 10 industries, AI referrals made up 1.08% of website traffic in Conductor's benchmark, and 87.4% of that came from ChatGPT (Conductor). ChatGPT adds a utm_source tag to many links, and Claude's web app passes a claude.ai referrer, but clicks from mobile apps, copied links and in-app browsers still land in Direct (Seer Interactive). Google also counts AI Overview clicks inside standard Search Console totals. A lot of AI-driven discovery gets filed as Direct or Organic.
  • Buyer intent is outsized. In July 2026, G2 reported that 51% of B2B software buyers now start their research by asking an AI chatbot, nearly double the 29% of just months earlier, and 69% chose a different vendor than they first had in mind because of what an AI told them (G2). Semrush modeled the average AI search visitor as worth 4.4 times a traditional organic visitor (Semrush), and on Ahrefs' own site, AI search was 0.5% of visitors and drove 12.1% of signups (Ahrefs).

Beyond Analytics: What Firewall and Edge Logs Reveal About AI Demand

If your agency is still hosting webinars about AI attribution while staring at referral traffic in Google Analytics, you're getting a rearview-mirror report.

Client-side analytics only measure human sessions where the tag is allowed to fire and the referrer survives. Privacy browsers and ad blockers stop the tag, mobile AI apps drop the referrer, and people research inside AI interfaces without clicking a link. To see what is actually happening, technical marketing teams go down a layer, to web application firewall (WAF), CDN and edge network logs.

The Pressfit.ai 30-Day Telemetry Check

Look at our own 30 days of telemetry at Pressfit.ai. Traditional web analytics showed about 6,500 human visits with almost no visible AI attribution:

  • Direct traffic: 6,460 visits (99.38%), absorbing dark social, AI app clicks with no referrer and cookie-blocked browsers.
  • Syndicated Google search: 20 visits (0.31%).
  • Grok referral: 10 visits (0.15%), the only AI assistant that showed up as a referrer in our report this month.
  • Paid social (Facebook ads): 10 visits (0.15%).

A surface-level marketer looking at that dashboard would conclude that ChatGPT, Claude and Meta AI send no traffic, and might question whether structured content for LLMs is worth the effort.

Then you open the firewall and network logs. Over the same 30 days, AI and search crawlers hit Pressfit.ai 17,630 times, 2.71 times our human visit count (22,692 requests including AdsBot and other Google services). AI and answer-engine bots alone made 7,321 requests, 1.13 times our human traffic.

The AI and Answer-Engine Layer: 7,321 Requests

Every one of these crawlers is an AI platform, retrieval API or autonomous agent reading Pressfit.ai, and almost none of them show up in Google Analytics.

AI provider30-day requestsShare of AI botsUser agents seen at the edgeRole in AI discovery
Meta AI2,28031.1%Meta-WebIndexer, Meta-ExternalAgentOur top AI crawler by volume, indexing and fetching live web answers for Meta AI across WhatsApp, Instagram and Facebook.
OpenAI / ChatGPT1,78924.4%OAI-SearchBot, GPTBot, ChatGPT-User, GPT Actions, OperatorSearch indexing (1,401) alongside live user-triggered lookups (127) and agent actions (129).
Amazon1,23416.9%AmazonbotFeeds Alexa answers, Rufus shopping retrieval and AWS Bedrock knowledge tools.
Parallel6138.4%ShapBotAI web search and retrieval infrastructure behind real-time agent APIs.
Anthropic / Claude5807.9%ClaudeBot, Claude-UserModel ingestion (523) plus live, human-prompted page fetches (57) during vendor research.
Exa4516.2%ExaSearchBot, ExaUserNeural search API behind developer-built AI apps, research agents and enterprise copilots.
Perplexity3464.7%PerplexityBotAnswer-engine indexing and citation retrieval.
Common Crawl170.2%CCBotOpen-web corpus used in open-source and enterprise LLM training.
DuckDuckGo AI110.2%DuckAssistBotPrivacy-first AI answers.

Break the traffic down by user agent and you stop asking "Did an LLM send a click?" and start measuring how often machines retrieve your pages:

  • Live, human-triggered fetches (ChatGPT-User, Claude-User, ExaUser). ChatGPT and Claude showed zero tracked referrals in our Google Analytics report, but our firewall logged 127 live ChatGPT-User fetches and 57 live Claude-User fetches on Pressfit.ai in the same 30 days. Those requests fire when a person asks a question that makes the model fetch our page to check facts or compare vendors.
  • Neural search infrastructure (Parallel, Exa). AI visibility isn't only about consumer chatbots. Your content is queried by neural search APIs (Exa: 451 requests; Parallel: 613) that sit underneath many vertical AI tools and enterprise copilots.
  • Autonomous agents (GPT Actions, Operator). Software agents are already querying pages (127 GPT Actions, 2 Operator sessions) and inspecting them before a person ever fills out a lead form.

The Search and Index Layer: 10,309 Requests

Traditional search crawlers still feed Google AI Overviews, Microsoft Copilot and Apple Intelligence:

Search crawler30-day requestsShare of search crawlersWhat it powers
Googlebot6,48662.9%Google Search index and Google AI Overviews grounding
Bingbot1,39313.5%Bing Search, Microsoft Copilot and third-party LLM search APIs
GoogleOther1,08710.5%Google research and specialized crawls
PetalBot7046.8%Huawei Petal Search and its mobile assistants
Baiduspider4053.9%Baidu Search and the ERNIE AI ecosystem
DuckDuckBot1051.0%DuckDuckGo search and private AI chat
Applebot790.8%Siri, Spotlight and Apple Intelligence answers
YandexBot500.5%Yandex Search and Alice AI

How to Measure Modern Search: The Attribution Stack

If your agency measures SEO and GEO only by referral traffic in an analytics dashboard, it is ignoring the thousands of machine retrieval events happening on your domain every month. Real SEO and GEO measurement connects four layers of data:

  • Deterministic content and site engineering. Structure entity relationships, schema and coverage of the follow-up questions behind each query, so crawlers can read your authority without friction.
  • Edge and firewall telemetry. Track user-agent volume across search indexers (Googlebot, Bingbot), model crawlers (Meta-WebIndexer, ClaudeBot), neural search APIs (Exa, Parallel) and live human-triggered fetches (ChatGPT-User, Claude-User).
  • Analytics and direct-lift correlation. Track the AI referrals you can see alongside spikes in unattributed Direct traffic on the exact landing pages getting live AI user fetches.
  • CRM and self-reported attribution. Ask inbound prospects how they found you before they fill out a form.

At Pressfit, we don't rely on guesswork, manual prompting or broken cookie reports. We engineer the deterministic playbooks that get your brand retrieved by AI engines, and connect the network, analytics and pipeline telemetry to prove it works.

Want the roadmap that comes before the measurement? Read The Pressfit Playbook: Your Roadmap to AI Visibility.

Need help putting it to work? Contact our team at redevteam@pressfit.ai.

See where you and your competitors rank in search and in AI answers.

Get your Playbook to outrank them. Paste your URL, and we do the hard work. Your Playbook arrives with your first report within 48 hours.

Get Onboarded

Newsletter

Prefer to read? Get new posts in your inbox — no spam, unsubscribe anytime.