AI search engines don’t return blue links, so your old SEO dashboards are now blind in one eye. You need a second layer of tracking covering four surfaces (Google AI Overviews, ChatGPT, Perplexity, Gemini) and six metrics (citation rate, share of citations, surface coverage, citation accuracy, AI referral traffic, branded prompt mentions).
Why “rankings” stopped being the right metric
In April 2026 we audited a B2B SaaS site that had been “winning” on Google for two years. Position 1–3 for its main category term. Steady CTR.
Then their CMO ran a single ChatGPT prompt: “What’s the best [category] for mid-market companies?”
The brand was not mentioned. Three competitors were. Two of those competitors ranked below them on Google.
That is the gap. “Ranking” tracks the SERP, but the SERP is no longer the only place buyers form their shortlist. ChatGPT alone reportedly serves hundreds of millions of weekly users in 2026, and Perplexity, Gemini and Google AI Overviews each take a slice of queries that used to be straight Google searches. If your tracking stops at SERP rank, you’re measuring a smaller and smaller surface every month.
The job now is to track whether your brand shows up when an AI system answers a question on your topic. That’s a different metric set, and most teams haven’t built it yet.
The four surfaces you need to track
There are dozens of AI surfaces, but four cover most B2B and B2C buyer behavior in 2026.
Google AI Overviews
The AI-generated answer block at the top of Google for many informational and commercial queries. Sources are cited inline. Your job: appear among the cited sources.
- Where it shows up: top of the SERP for AI-eligible queries.
- What gets cited: clean, well-structured pages with direct answers, recent dates and recognizable domains.
- Tracking signal: does your URL appear in the citations panel for your priority queries?
ChatGPT (Search and chat)
Two surfaces inside one product: standard chat (from the model’s training plus its retrieval layer) and ChatGPT Search (real-time web).
- Standard chat draws from training data plus a retrieval layer, which means citations can come from sites the model has been exposed to broadly.
- ChatGPT Search behaves more like a live search engine and cites pages directly.
- Tracking signal: does ChatGPT name your brand, link your domain, or both, in answers to relevant prompts?
Perplexity
The most “search-engine-like” of the AI answer engines. Every answer is built around live citations, and links to sources are prominent.
- The fastest surface to move on with good content: Perplexity surfaces sources aggressively.
- Tracking signal: are you cited in the source panel for your priority prompts?
Gemini and Google’s other AI surfaces
Google Gemini, AI mode in Search, and AI summaries in Discover. Citation behavior varies by surface and is the least transparent of the four.
- Tracking signal: presence in cited sources within Gemini answers.
The 6 metrics that actually matter
1. Citation rate
For a defined list of 30–100 priority prompts, what percentage of answers include your domain as a cited source? Track per surface and combined.
Example baseline (real-ish): Citation rate on Perplexity = 22% (11 of 50 prompts). On Google AI Overviews = 7%. Goal Q3: lift Overviews to 15%.
2. Share of citations vs competitors
Of all citations across your prompt set, what share is your domain vs each named competitor? This is the AI-era equivalent of share of voice.
3. Surface coverage
Of the four major surfaces, on how many are you cited at all? A brand cited only on Perplexity has fundamentally different exposure than one cited across all four.
4. Citation accuracy
When AI answers describe your brand, are they describing it correctly? An AI that confidently says you do X when you actually do Y is worse than not being cited at all.
- How to test: prompt “What does [your brand] do?” across all surfaces. Score the answer for accuracy.
5. Referral traffic from AI sources
In your analytics, how much traffic is arriving from chatgpt.com, perplexity.ai, gemini.google.com, bing.com, plus the long tail of AI assistant referers? Tag these in GA4 as a custom channel.
6. Branded prompt mentions
For prompts that don’t mention you, does your brand still get named in the answer? This is the leading indicator of category presence in AI search.
Tools we’ve used (and what they actually do)
We’ve used or tested these in client work and on our own sites between January and May 2026. Honest take on each:
Profound
- What it does: tracks brand mentions and citations across major AI surfaces, with competitive benchmarking.
- Strength: enterprise-grade dashboard, broad surface coverage.
- Weakness: pricing is enterprise-grade too.
- Best for: in-house marketing teams at companies with budget and at least 3–5 prompt categories worth tracking.
Otterly.ai
- What it does: monitors brand and prompt visibility across ChatGPT, Perplexity and Google AI Overviews, alerts on changes.
- Strength: easier to onboard than enterprise tools, fair pricing for SMB.
- Weakness: surface coverage and prompt limits depend on plan.
- Best for: agencies and growth teams running multiple smaller accounts.
Peec AI
- What it does: similar category, with a focus on competitive share-of-voice across LLM answers.
- Strength: clear competitive view.
- Best for: brands in saturated categories where the question is “are we losing share to X?”
HubSpot AI Search Grader
- What it does: free tool that scores how visible your brand is in AI search.
- Strength: free, fast, decent first signal.
- Weakness: it’s a grader, not a tracker. Use it as a starting point, not as your dashboard.
SEMrush AI Toolkit
- What it does: AI visibility module inside SEMrush’s broader stack.
- Strength: useful if you already pay for SEMrush.
- Weakness: depth varies by plan and the feature is still evolving.
Manual tracking (the free version)
If you don’t have budget yet, do this:
- Pick 30 priority prompts that real buyers type.
- Every two weeks, run them across all four surfaces from a clean session (logged out, no personalization, VPN off).
- Log the answer text, the citations and whether your brand is named.
- Track in a Google Sheet with columns: prompt, surface, date, brand mentioned (Y/N), domain cited (Y/N), accuracy score 1–5, notes.
Manual tracking is tedious but produces the cleanest baseline. The paid tools save time once you know what to measure.
How to set up tracking in one afternoon
- Define your prompt set (90 minutes). Pull 30–100 prompts from real sources: support tickets, sales call recordings, search console queries, the “people also ask” panel for your top SERPs, your competitors’ FAQ pages.
- Pick your tooling tier (15 minutes). Free + manual, mid-tier (Otterly), or enterprise (Profound). Don’t overcomplicate this on day one.
- Establish a baseline (2 hours). Run every prompt across all four surfaces. Capture screenshots and the answer text.
- Build the dashboard (1 hour). Six metrics, one row per prompt, one column per surface, plus a summary tab.
- Schedule a re-run (5 minutes). Bi-weekly is enough for most categories. Weekly only if the category is moving fast (AI tooling itself, anything regulatory).
A 30-day baseline you can present to a CMO
- Total prompts tracked: 50.
- Citation rate per surface (start vs target).
- Top 3 prompts where you’re cited.
- Top 3 prompts where you’re missing.
- Share of citations vs your top 3 competitors.
- One slide called “where competitors are eating us” with named prompts and named competitors.
- One slide called “what we’ll do in the next 30 days” with a ranked list of pages to fix or create.
That’s enough to justify the work. More dashboard than that on day one is theater.
What to do when you’re not getting cited
The fixes break into three buckets, in this order:
- Page-level: direct answer in the first 80 words, FAQPage schema, recent dateModified, named author with credentials, outbound citations to primary sources. See our GEO for Startups playbook for the page-level checklist.
- Site-level: about page, editorial policy, author bios, consistent NAP and entity data, organization schema, robots.txt that allows GPTBot, ClaudeBot and PerplexityBot.
- Off-site: third-party citations from sites the model already trusts (Wikipedia, Reddit threads with real engagement, industry publications). Without this, page-level fixes have a ceiling.
What’s broken in the current tooling (be honest)
Three caveats no vendor will tell you:
- AI answers are partially personalized and partially probabilistic. The same prompt at the same time can return different answers. Single runs are noisy. Track trends over weeks, not single data points.
- “Visibility” tools differ in how they count mentions, citations and brand inclusions. Comparing across tools is unreliable.
- Some surfaces (notably Gemini) provide weaker citation data than others. Coverage is uneven by design.
You can still build a useful program around this. Just don’t pretend the numbers are as crisp as Google rank tracking was in 2018.
Frequently Asked Questions About Slow Business
What’s the difference between GEO and AI search tracking?
GEO is the optimization work making your site interpretable, credible and citable to LLM-based engines. AI search tracking is the measurement layer that tells you whether the optimization is working.
How often should I track AI search visibility?
Bi-weekly for most categories. Weekly if your category is moving fast or you’re running an active GEO sprint.
Can I rely on Google Search Console for this?
Not for the AI surfaces. GSC reports impressions and clicks for Google Search, including some AI Overview attribution, but it does not tell you anything about ChatGPT, Perplexity or Gemini.
Is there a free way to do this well enough to start?
Yes manual tracking on 30 prompts, once every two weeks, in a Google Sheet. It’s how we ran the first month of every client engagement before paid tools were mature.
How do I know if the AI is describing my brand accurately?
Run a “brand truth” prompt: “What does [brand] do?” across all four surfaces. Score answers 1–5 for accuracy. If you score under 4, your About page, homepage and structured data are sending mixed signals fix those before chasing more visibility.
