Semrush for GEO: Tracking AI Visibility (Setup + Interpretation)

TL;DR

Use the AI Visibility Toolkit for prompt-level brand mentions and competitor share of voice, and use Position Tracking for keyword-level rankings in Google AI Mode and ChatGPT. Pair them with the AI Overviews filter in Organic Research to see where your existing pages already get cited. Treat Semrush data as a directional signal, not a literal log of LLM responses — re-verify high-stakes prompts manually.

What "AI visibility" means in Semrush

Semrush splits AI visibility into three measurable layers:

• Mentions — how often your brand name appears inside an AI answer for a tracked prompt.
• Citations — how often your domain is linked as a source under an AI answer (most relevant for Google AI Overviews and Perplexity).
• Position / Rank — for AI surfaces that return ranked results (Google AI Mode, ChatGPT search), the slot your URL occupies.

These three numbers are aggregated into a single AI Visibility Score at the brand level, and into per-prompt scores inside the toolkit. The data comes from Semrush's Prompt Database, a panel of LLM responses Semrush queries on a recurring schedule, not a live API call into ChatGPT each time you load the dashboard.

Why GEO teams use Semrush specifically

Three reasons stand out for teams already running SEO inside Semrush:

1. One project, two layers of search. Semrush One unifies the SEO Toolkit and the AI Visibility Toolkit, so you can compare classic organic rankings and AI mentions against the same competitor set without exporting CSVs.
2. Coverage across the four AI surfaces that matter. ChatGPT, Perplexity, Gemini, and Google AI Overviews / AI Mode are all tracked from the same dashboard.
3. Competitor benchmarking by default. The Competitor Research report compares mentions and citations against up to four competitors side by side, which is what most stakeholders actually ask for.

If you do not already use Semrush, point tools like Profound, Evertune, or Otterly may have a lower price floor — but they will not deduplicate against your existing keyword data the way Semrush One does.

Prerequisites

Before you start tracking, gather:

• A Semrush One, Business, or higher plan (Position Tracking on ChatGPT and Google AI Mode requires Guru tier or above; the AI Visibility Toolkit ships with Semrush One).
• A list of 15-50 prompts representing how customers describe the problem your product solves. Pure brand-name lookups are fine but they under-represent discovery prompts.
• A confirmed brand name + 1-3 brand aliases (legal entity, common misspelling, product line). Semrush deduplicates mentions on the canonical brand string.
• A short competitor list (up to four) you want benchmarked.

Step-by-step setup

Step 1: Create the AI Visibility project

From the Semrush dashboard, open AI Visibility Toolkit → Projects → New project. Add your domain, brand name, and aliases. Pick the AI platforms you want tracked (ChatGPT, Perplexity, Gemini, Google AI Overviews are typical defaults). Set the market (country + language) — visibility scores differ sharply between markets, so choose the one your customers actually search from.

Step 2: Add prompts

You can populate prompts three ways:

• Manual paste. Paste your 15-50 customer prompts directly. Best when you already have voice-of-customer data.
• Prompt suggestions. Semrush proposes prompts based on your domain and topic clusters. Useful as a top-up, but review for relevance before saving — the suggestion engine pulls from generic intent patterns and can include off-target phrasings.
• Import from keywords. Convert tracked SEO keywords into question-form prompts. Helpful for showing how classic head terms translate into LLM queries.

Aim for a balanced prompt mix: roughly 30% branded, 50% category / problem-led ("best CRM for solo founders"), 20% comparison ("X vs Y").

Step 3: Add competitors

Add up to four competitors in Competitor Research. Competitors must be domains, not brand strings. If a competitor operates multiple TLDs (.com, .co.uk), pick the one that matches your tracked market.

Step 4: Connect Position Tracking for Google AI Mode and ChatGPT

Position Tracking is a separate tool but consumes the same prompt list. Open Position Tracking → New campaign → Search engine: Google AI Mode (or ChatGPT). Paste the same prompts so the keyword-level ranks line up with the prompt-level mentions in the AI Visibility Toolkit.

Step 5: Wait one full collection cycle

Semrush refreshes prompt-level data daily for priority prompts and weekly for the broader prompt list, depending on tier. Do not interpret day-zero numbers as a baseline — wait at least 7 days of stable data before drawing conclusions.

Reading the dashboards

Visibility Overview

This is the rollup you screenshot for stakeholder updates. Three numbers matter:

• AI Visibility Score — your weighted share of mentions + citations across all tracked prompts, per platform.
• Mentions trend — week-over-week change in absolute mentions. A drop with no content change usually means a platform updated its model or shifted its source mix.
• Share of Voice — your mentions as a percentage of mentions for you + the four tracked competitors.

Mentions and Citations

Drill in by platform. For each prompt you can see:

• The answer snippet Semrush captured.
• The sources cited under the answer.
• The position of your domain in the source list (if any).

Use this view to verify why a score moved. A common pattern: your mentions stay flat but a competitor's mentions spike because a new third-party listicle started getting cited by ChatGPT.

Competitor Research

The gap-analysis view. Filter by "prompts where competitor is mentioned and you are not" — this is the highest-leverage backlog for the content team. Export to CSV and tag each row with the underlying intent and the page (existing or new) that should answer it.

Position Tracking — Google AI Mode and ChatGPT

For each tracked prompt:

• AI Mode position — the ranked slot your URL occupies, if any.
• AI Mode mention — boolean for whether your brand name was named in the answer body.
• Visibility % — a normalized score across the prompt set.

A URL can have a mention without a position (Semrush picked up your brand name in prose but no link to your site) and vice versa. Both count, but mention-without-position is more fragile because it depends on the model continuing to recall the name.

AI Overviews via Organic Research

For pages already ranking in Google, open Organic Research → Positions → SERP Features → AI Overview → Domain Ranks. This shows which of your pages are already cited inside AI Overviews. Mirror it with Domain doesn't rank to find pages that rank organically but never get pulled into the AI Overview — these are your fastest-to-fix candidates.

How often to check

• Daily: only for crisis-tracking (a launch, a takedown, a viral competitor mention).
• Weekly: Visibility Overview, Mentions trend, Competitor gap export.
• Monthly: Position Tracking comparison, AI Overviews coverage on your top 50 pages.
• Quarterly: Re-validate the prompt list. Customer language drifts faster in AI search than in classic SEO, so prune dead prompts and add 5-10 new ones each quarter.

Common data pitfalls

1. Treating Semrush as a literal LLM log

Semrush queries the platforms on a schedule from its own panel. The exact answer your customer sees in their session can differ because of personalization, conversation history, and model version. Re-run high-stakes prompts manually in ChatGPT, Perplexity, and Gemini before quoting numbers in a board deck.

2. Confusing mentions with citations

A mention is a brand-name appearance. A citation is a clickable source link. Many GEO dashboards (Semrush included) show both, and stakeholders often conflate them. Citations drive referral traffic; mentions drive recall. They are not interchangeable KPIs.

3. Reading day-one drops as regressions

AI surfaces are noisy. Single-day swings of 10-20% in Visibility Score are normal even with no content changes. Use 7-day rolling averages (available as a chart toggle) before declaring a trend.

4. Tracking only branded prompts

Branded prompts will look great because the model already knows your name. The prompts that move revenue are unbranded category and comparison prompts where the model is choosing whom to mention.

5. Ignoring market and language splits

A prompt tracked in en-US and the same prompt tracked in en-GB can produce different mention counts because the model retrieves different sources. Pick the markets that match your buyer geography and track them as separate projects.

What good looks like after 90 days

A team using Semrush AI visibility tracking productively will, by the end of a quarter, have:

• A stable prompt list of 30-50 prompts with at least 7 days of data each.
• A weekly cadence of reading Visibility Overview and exporting the competitor gap CSV.
• A backlog of content briefs derived from "competitor mentioned, you are not" prompts.
• Two dashboards in My Reports: one for executives (Visibility Score + Share of Voice), one for the content team (per-prompt mentions + citation list).
• A documented manual spot-check process for high-stakes prompts.

FAQ

Q: Does Semrush track ChatGPT in real time?

No. Semrush queries ChatGPT and other LLMs on a recurring schedule from its own prompt panel and stores the results in the Prompt Database. Priority prompts refresh daily, broader prompts refresh weekly. Always re-run high-stakes prompts manually before quoting numbers.

Q: What is the difference between the AI Visibility Toolkit and Position Tracking for AI?

The AI Visibility Toolkit reports prompt-level mentions and citations across ChatGPT, Perplexity, Gemini, and Google AI Overviews. Position Tracking reports keyword-level ranked positions inside Google AI Mode and ChatGPT search. Use the toolkit for share-of-voice reporting and Position Tracking for SERP-style rank tracking.

Q: Which Semrush plan do I need for AI visibility tracking?

The AI Visibility Toolkit ships with Semrush One. Position Tracking on ChatGPT and Google AI Mode requires the Guru tier of the SEO Toolkit or above. Free trials cover both, but limit prompt counts and competitor slots.

Q: How many prompts should I track?

Start with 15-50 prompts, balanced roughly 30% branded, 50% category, 20% comparison. Scale up only after 90 days of stable data — extra prompts dilute attention more than they add insight.

Q: Can I trust the AI Visibility Score as a single KPI?

Treat it as directional. The score is a weighted blend of mentions, citations, and rank, and the weighting is opaque. Report it alongside raw mentions and citations counts so stakeholders see the underlying movement, not just the composite.

Q: Why do my mentions drop overnight with no content change?

Usually a model update on the platform side, or a shift in which third-party sources the model is retrieving. Wait for 7 days of rolling data before treating it as a regression, and check whether competitor mentions also moved on the same day — a market-wide shift points to a platform change rather than a content problem.