Measure AI Search Against Pipeline With Ahrefs Brand Radar

AI-search reporting is drifting toward the same mistake SEO made in 2013: counting visibility because it is countable.

Ahrefs Brand Radar makes the count part much easier. As of 2026, Ahrefs says Brand Radar tracks AI visibility across 461M+ search-backed prompts in its AI Visibility Index, with coverage for Google AI Overviews, AI Mode, ChatGPT, Perplexity, Gemini, and Copilot. Its help docs also describe metrics like mentions, citations, impressions, and AI Share of Voice. That is useful raw material. It is not a board slide by itself.

I would treat Brand Radar as a source table, not a KPI layer. The operating question is narrower: when our brand shows up in AI answers for commercial prompts, do we see more qualified key events in GA4 and more sourced or influenced pipeline in HubSpot or Salesforce within the next 30 to 90 days?

That framing matters because a mention in ChatGPT is not the same thing as demand. A B2B founder running paid search, LinkedIn ads, partner content, and comparison pages does not need one more dashboard showing that the brand appeared 428 times last month. She needs to know whether those appearances clustered around prompts that buyers actually use before booking a demo, starting a trial, or asking procurement for a security review.

Start With Prompts That Smell Like Pipeline

The first pass is always prompt hygiene. Ahrefs says Brand Radar models prompts from real search behavior, including Google’s People Also Ask data and semantic fanout from its keyword database of more than 100 billion keywords. That gives you breadth. Your job is to cut it down to buyer intent.

For a security compliance startup, I would separate “what is SOC 2” from “best SOC 2 automation tools for startups” and “Vanta vs Drata pricing.” The first prompt belongs in education reporting. The second and third belong in pipeline reporting. Same tool, different business meaning.

In Brand Radar, I would create saved reports for three prompt groups: category, alternatives, and pain. Category prompts include phrases like “best customer data platform for B2B SaaS”. Alternatives prompts include “Segment alternatives” or “RudderStack vs Hightouch”. Pain prompts sound like a buyer who has hit a wall, such as “how to sync product usage data to Salesforce without engineering tickets.” That last one is messy. Good. Messy prompts are where buying committees leak intent.

Ahrefs Custom Prompts are useful here because they track exact buyer questions, refreshed daily on selected platforms according to the Brand Radar product page. If you sell into a small niche, the broad AI Visibility Index may miss the wording your prospects use on calls. I keep a tiny prompt library from sales notes: 40 to 80 questions pulled from Gong, Chorus, HubSpot call notes, and demo forms. That set beats a clean spreadsheet full of generic category phrases.

Use Brand Radar Metrics As Inputs

Brand Radar’s mention metric counts an AI response once when the brand appears in that response, even if the brand name appears three times. Citations count when a page is cited as a source. Ahrefs also tracks “found in” pages, which can include pages retrieved in the background but not cited in the answer. That distinction matters.

A cited pricing page has a different value from a background retrieval of a glossary post. If Perplexity cites your “SOC 2 automation pricing” page in 19 commercial responses during August 2026, I want that next to GA4 key events from the same landing page family. If ChatGPT retrieves a five-year-old explainer but never names the company in the response, I would still log it, but I would not let the marketing team celebrate it in the Monday pipeline meeting.

Build the Brand Radar extract at the weekly grain. You need date, platform, prompt, prompt group, brand mentioned, competitor mentioned, cited URL, cited domain, found URL, estimated impressions, and share of voice. If Ahrefs API access is available on your plan, pull it into BigQuery. If not, start with scheduled CSV exports and stop pretending manual data is beneath you. A clean weekly export every Monday at 8:00 a.m. Eastern is enough for the first version.

I would also add two hand-coded fields before the data touches Looker Studio: intent tier and owned asset type. Intent tier can be 1, 2, or 3. Tier 1 is bottom-funnel comparison, pricing, demo, migration, or vendor-selection language. Tier 2 is category and solution research. Tier 3 is education. Owned asset type is pricing page, comparison page, integration page, blog post, docs, free tool, partner page, or other.

GA4 Key Events Are The Behavior Layer

Google renamed Analytics conversions to “key events” in 2024 so Analytics and Google Ads would stop using the same word for slightly different measurement concepts. The official Google Analytics help page defines a key event as an event that measures an action important to the business. Google also added key event fields to the GA4 Data API on May 6, 2024, including keyEvents, sessionKeyEventRate, and isKeyEvent.

For this use case, I want three GA4 key events, not twelve. On a B2B site, that usually means generate_lead, sign_up, and book_demo. If you have a pricing-page CTA that opens a HubSpot meeting router, mark the final booking event as the key event, not the button click. Button clicks are useful debugging exhaust. They are not pipeline.

The practical join is not perfect because AI answers often create dark traffic. Someone asks Perplexity for “best SOC 2 tools for startups,” sees your name, then types the domain directly two days later. GA4 will call that Direct, Organic Search, Paid Search, or whatever the last measurable touch says. Fine. Do not force fake precision.

Use a date-window model first. For each prompt group and platform, compare Brand Radar visibility in week N with GA4 key events in weeks N, N+1, and N+2 for matching page families. If AI citations to comparison pages rose from 12 to 41 in August, and demo bookings from comparison pages moved from 18 to 29 over the next two weeks while spend stayed flat, you have a thread worth pulling.

Looker Studio can blend up to five data sources, according to Google’s docs, but I would avoid doing heavy logic inside blends. Use BigQuery, dbt, or even a controlled Google Sheet to shape the data first. Looker Studio should show the model, not become the model.

The CRM Join Is Where The Dashboard Grows Up

GA4 tells you behavior. The CRM tells you whether sales accepted the behavior as money-shaped.

In Salesforce, the simplest version uses Campaign Member, Lead Source, Primary Campaign Source, Opportunity Amount, Stage, Created Date, and Close Date. Salesforce’s own help docs point teams to Primary Campaign Source and Campaign Revenue reports for tracking business generated by campaigns. In HubSpot, deals have properties like deal stage, amount, close date, closed won status, and recurring revenue fields on the right editions. Those are the columns that matter.

I like creating an “AI search influence” campaign in the CRM, but only with strict rules. A contact gets influence when three things happen: the first or recent session lands on an AI-cited page family, the session occurs within 14 days of a Brand Radar mention or citation for the relevant prompt group, and the contact creates a key event such as book_demo or generate_lead. That is attribution with dirt under its nails. It will miss some deals. It will also avoid claiming every direct lead that happened after your brand showed up somewhere in Gemini.

For a real operating dashboard, I would show four rows by prompt group: AI Share of Voice, cited owned pages, GA4 key events, and CRM pipeline. Then I would add one ratio that makes people uncomfortable: pipeline per 100 AI mentions. If category prompts generated 1,200 mentions and $18,000 in created pipeline, while alternatives prompts generated 140 mentions and $96,000, the work is obvious. Build and earn citations for alternatives content before writing another top-of-funnel guide.

A plausible benchmark from one SaaS account I would expect to see in Q4 2026: 9 percent AI Share of Voice on category prompts, 22 percent on alternatives prompts, 64 demo bookings from AI-cited page families, 17 sales-qualified opportunities, and $310,000 in created pipeline. The exact numbers will move by ACV and sales cycle. The shape is what you are watching.

What I Put In Looker Studio

The first page is an executive view, but keep it spare. Brand Radar visibility goes on the left. Pipeline goes on the right. The middle shows the bridge: key events from GA4.

Use scorecards for AI Share of Voice, owned citations, key events, SQLs, created pipeline, and closed-won revenue. Under that, add a time series with weekly Brand Radar mentions and GA4 key events. The second chart should lag key events by seven and fourteen days, because same-week correlation will lie to you on considered purchases. A founder looking at a 45-day sales cycle needs lag, not instant gratification.

The second page is for operators. Break it down by platform, prompt group, competitor, and cited URL. I want to see that Perplexity cited our comparison page 31 times, Google AI Overviews cited G2 and Capterra instead, and ChatGPT mentioned two competitors without citing anyone. That points to different work. Perplexity may reward source-page cleanup. Google may require third-party proof. ChatGPT may require broader category association through docs, partner pages, and review sites.

The third page is a QA page. Put raw rows there: prompt, response date, platform, brand mentioned, cited URL, GA4 landing page, key event count, CRM opportunity ID, and amount. Every serious marketing dashboard needs a place where someone can click into the weird row and decide whether the model is lying.

Do The Boring Naming Work

Most AI-search dashboards fail because nobody standardizes names. Brand Radar may show a cited URL with UTM-free canonical paths. GA4 may store landing pages with query strings. HubSpot may have lifecycle stages that sales changed in March. Salesforce campaign names may include “FY26-Q1” in one region and “Q1 FY2026” in another.

Normalize before reporting. Strip URL parameters except the ones you deliberately keep, usually utm_campaign, utm_source, utm_medium, and a custom ai_prompt_group if you use one. Map /compare/acme-vs-contoso, /alternatives/contoso, and /blog/contoso-alternatives into the same comparison family when that is how buyers experience the journey. Use lowercase brand names for joins. Keep a competitor alias table with entries like “Google Analytics”, “GA4”, and “Google Analytics 4”.

I would also keep a manual override table. Ten rows will save you ten hours. When a CRM campaign was renamed, when a product page moved, when an acquisition changed a brand name, when a partner directory starts outranking your site, write it down as data. Operators forget. Tables don’t.

The Cadence I Trust

Monthly is too slow at the start. Daily is theater for most B2B teams. I would run the operating review weekly for the first 8 weeks, then move to twice a month once the model stops surprising you.

The meeting should answer three questions. Which prompt groups created pipeline per mention? Which cited pages drove key events? Which competitors gained AI Share of Voice without gaining pipeline? That last one keeps the team honest. A competitor can win mentions on education prompts and still lose buyers at the demo step.

After 90 days, decide where the work goes. If AI Overview citations are coming from G2, Capterra, Reddit threads, and integration docs, your SEO backlog changes. If ChatGPT mentions competitors because their category pages explain use cases better than yours, rewrite the pages buyers actually touch. If Perplexity cites your docs but prospects bounce before the demo form, fix the path from docs to sales.

Brand Radar tells you where the AI layer sees you. GA4 tells you what people did next. Salesforce or HubSpot tells you whether sales could turn that behavior into revenue. Keep those three layers joined, and AI-search reporting becomes an operating system instead of another impressions chart.