How to Track When ChatGPT, Perplexity, and Google AI Overviews Cite Your Brand

A founder types "best mineral sunscreen for kids" into ChatGPT. Her competitor gets named twice. Her own store ranks on page one of Google for that phrase, but it isn't mentioned at all. She checks again the next morning, and her brand shows up third. Tracking brand citations in AI answer engines usually starts like this: confusing, inconsistent, and easy to misread.

It's worth getting right anyway. Seer Interactive (Nov 2025) found organic CTR on informational queries fell 61% when an AI Overview appeared. Over the same stretch, Semrush (2026) reports ChatGPT's outbound referral traffic to the web grew 206% in 2025. The answer box is turning into a second traffic channel.

What You'll Build

You'll finish with four things:

• A fixed prompt list.
• A weekly citation log covering ChatGPT, Perplexity, and Google AI Overviews.
• A GA4 channel that captures AI referrals.
• One weekly view that puts each article's AI citations next to its Search Console clicks.

Before You Start

• Search Console: a verified property with at least 3 months of data.
• GA4: Editor access or higher. You need it to create custom channel groups.
• AI accounts: ChatGPT (the free tier works if answers show sources), Perplexity (the free tier is fine), and a Google account.
• Spreadsheet: Google Sheets or Excel. The formulas below use Google Sheets syntax.
• Target URLs: a list of 10–20 published articles you most want cited.
• Time: about 90 minutes for setup. Manual checks then take 2–3 hours per week.
• Skill level: you can filter a Search Console report and change a GA4 setting.

Step 1: Tracking Brand Citations in AI Answer Engines Starts With a Fixed Prompt List

AI answers shift with small changes in wording. You need to ask the same questions every week. A fixed list turns random spot checks into data you can compare over time. Aim for 20 to 30 prompts that match how your buyers actually ask.

1. Open Search Console and go to Performance > Search results.
2. Click + New > Query > Custom (regex). Then paste this filter:

^(how|what|why|which|best|is|can|should)\b

3. Sort by impressions. Copy the top 10 question-style queries.
4. Rewrite each one as a full sentence a person would type into ChatGPT. For example, "best kids sunscreen mineral" becomes "What's the best mineral sunscreen for a 5-year-old with sensitive skin?"
5. Add 5 comparison prompts, such as "Is Brand A or Brand B better for sensitive skin?"
6. Add 5 recommendation prompts that leave out your brand name, such as "Which small sunscreen brands use reef-safe ingredients?"
7. Save everything in a tab called Prompts. Use these columns: prompt_id, prompt, intent, target_url.

✅ Check: Your Prompts tab has 20–30 rows. Every row maps to one URL you'd want an AI engine to cite.

Step 2: Manual Checks Only Mean Something When You Repeat Them

A single run of a prompt tells you very little. SparkToro and Gumshoe.ai (Jan 2026) found that ChatGPT gave the same brand list twice less than 1 time in 100. You'll run each prompt in all three engines, on three separate days each week.

1. Open a private or incognito window for every session. Logged-in history and memory skew the results.
2. ChatGPT: start a Temporary Chat and paste the prompt. Confirm the answer lists sources. If it doesn't, ask it to search the web.
3. Perplexity: paste the prompt in the default mode. Then open the Sources panel.
4. Google: search the prompt in incognito. Note whether an AI Overview appears. If one does, expand it and check the linked sources.
5. Log every result in a tab called Log with these columns:

date | engine | prompt_id | brand_named (Y/N) | our_url_cited (Y/N) | cited_url | competitors_named

6. Repeat the full run on two more days that week.

Don't record a "position." Rand Fishkin of SparkToro argues that AI recommendations shift almost every time you run a prompt. In his view, any tool reporting a rank position in AI answers gives a misleading number.

You need all three engines because they rarely agree. A vendor study by Mentions.ai (Jul 2026) found that 69% of sources behind a typical answer changed overnight. The same study found 84% of cited sources were used by just one engine.

✅ Check: After one week, you have roughly 225 rows (25 prompts × 3 engines × 3 days). Each engine has about 75 rows.

Step 3: A Mention and a Citation Are Different Results

An engine can name your brand without linking to you. It can also use your content without naming you. Seer Interactive's March 2026 work calls the second case "ghost citations." Splitting these outcomes shows which articles earn links and which only earn name-drops.

| Outcome | Brand named | Your URL linked | What it usually means | |---|---|---|---| | Full citation | Yes | Yes | Your page is the source and you get credit | | Mention only | Yes | No | The engine knows you from third-party sources | | Ghost citation | No | Yes | Your content is used, but your brand isn't credited | | Absent | No | No | You're not part of this answer |

1. Add an outcome column (column H) to the Log tab. Paste this in H2 and fill it down:

=IFS(AND(D2="Y",E2="Y"),"Full",D2="Y","Mention",E2="Y","Ghost",TRUE,"Absent")

2. Create a pivot table. Set rows to engine, columns to outcome, and values to COUNTA of prompt_id.
3. Filter for "Mention" rows. Read the cited_url column to see which outside sites feed the engine's view of you.

That third action matters. Ahrefs (2026) studied 75,000 brands and found YouTube mentions had the strongest correlation with AI visibility (about 0.737). Branded web mentions also correlated highly (0.66–0.71). Mention-only rows often point to Reddit threads, YouTube reviews, or listicles you could appear in.

✅ Check: The pivot shows four outcome counts per engine. Each engine's counts add up to its total rows.

Step 4: GA4 Needs a Custom Channel for AI Referrals

Citations matter most when they send visitors. GA4 added a native AI Assistant channel on May 13, 2026. DarwinApps (2026) notes it skips Perplexity and every session that arrives without a referrer. A custom channel group closes part of that gap.

1. In GA4, go to Admin > Data display > Channel groups.
2. Click Create new channel group. Name it "Default + AI."
3. Click Add new channel. Name it "AI Answer Engines."
4. Set the condition to Session source, matches regex, and paste:

chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai

5. Drag the new channel above Referral. GA4 checks channels from top to bottom.
6. Save. Custom channel groups apply to historical data too.
7. Open Reports > Acquisition > Traffic acquisition. Switch the primary dimension to your new channel group.

One limit applies here. Clicks from Google AI Overviews arrive as google / organic, so GA4 can't separate them from regular search clicks.

✅ Check: "AI Answer Engines" appears as a row in Traffic acquisition. If you get any Perplexity visits, they show up inside it.

Step 5: Search Console Covers the Google Side

Clicks from Google AI Overviews land inside your regular organic data. That makes Search Console your main source for them. Google launched a Generative AI performance report on June 3, 2026. It went first to a subset of UK sites and reports impressions, not clicks.

1. Open Search Console and check the left menu for a Generative AI report. If you have it, export impressions by page.
2. If you don't have it, use Performance > Search results.
3. Set the date range to Compare > Last 28 days vs. previous period.
4. Filter Page by each target URL. Note clicks, impressions, CTR, and position.
5. Look for pages where impressions rose but clicks stayed flat or fell. That pattern often means an AI Overview is answering the query before anyone clicks. Ahrefs data from December 2025 showed AI Overviews cut CTR for top-ranking results by around 58%.
6. Export the page data to a tab called GSC.

✅ Check: Your GSC tab has one row per target URL. Each row shows 28-day clicks and impressions.

Step 6: One Weekly View Joins Citations, Referrals, and Clicks

Each source shows only one slice. Search Console covers Google, GA4 covers referrals it can identify, and your log covers a sample of answers. Joining them per article shows which pages get cited and whether those citations turn into visits.

| Source | What it shows | What it misses | |---|---|---| | Search Console | Google clicks and impressions per page | Which clicks came from AI Overviews (outside the UK pilot) | | GA4 custom channel | Sessions from ChatGPT, Perplexity, Gemini | Referrer-less visits and AI Overview clicks | | Citation log or tracker | Whether engines name or link you | Anything outside your sampled prompts |

1. Create a tab called Weekly with one row per target URL in column A.
2. Add the citation rate in column B. The 9 equals 3 days × 3 engines:

=COUNTIF(Log!F:F,A2)/(COUNTIF(Prompts!D:D,A2)*9)

3. Pull GSC clicks into column C:

=XLOOKUP(A2,GSC!A:A,GSC!B:B,0)

4. Export GA4 landing pages filtered to "AI Answer Engines." Add those sessions in column D.
5. Judge trends on 4-week rolling averages, not single weeks.

The manual version works, but it costs 2–3 hours a week and still only samples. Paid tools cut that time. Otterly.AI starts at $29/month. Semrush's AI Visibility Toolkit tracks AI Overviews and ChatGPT mentions next to rankings. Ahrefs Brand Radar stores AI Overview responses and searches them for brand mentions. Profound serves enterprise brands like Indeed, MongoDB, and Ramp.

CocoSEO handles tracking brand citations in AI answer engines as part of its publishing loop. It reports which of your articles ChatGPT, Perplexity, and AI Overviews cite, next to each article's Search Console clicks. Those numbers arrive in a weekly digest.

✅ Check: Sort the Weekly tab by citation rate. Your top 5 cited articles appear with their clicks and AI sessions beside them.

Troubleshooting

Your brand appears one day and vanishes the next

A brand that appears in an AI answer one day and vanishes the next is normal behavior, not a broken setup. AI answers from ChatGPT, Perplexity, and Google AI Overviews change between runs, so the same prompt can return different sources and different brand mentions from one check to the next.

Because of that run-to-run variation, a single check is a poor measure of visibility. Track the share of runs where your brand appears over a 4-week window. Don't react to any single check, whether the result is a new mention or a missing one. The 4-week share is the number to watch.

GA4 shows almost no AI traffic

When GA4 shows almost no AI traffic, the AI visits are usually being misattributed rather than missing. Many AI apps strip the referrer, so visits from AI answer engines land in GA4's Direct channel instead of Referral or a custom AI channel. As a result, GA4 reports fewer AI visits than actually occurred.

To check your GA4 setup, confirm your custom AI channel sits above Referral in the channel group order. Then check for a rise in Direct traffic to your target URLs, since stripped-referrer AI visits will show up there. Expect GA4 to undercount AI visits even with a correct setup.

Two tools report very different numbers

Two AI visibility tools report very different numbers because each tracker samples different prompts, regions, and models. Each tool is effectively measuring a different slice of AI answers, so brand mention counts from two trackers are rarely comparable, even for the same brand over the same period.

The gap between trackers can be large. One review found Ahrefs' Brand Radar reported 3 ChatGPT mentions for a brand that a manual check put at 123. Writesonic, which sells a competing tool, published that review, so treat the Brand Radar comparison as anecdotal. Pick one tool and compare that tool against itself over time.

Search Console has no Generative AI report

A missing Generative AI report in Google Search Console is expected for many sites. Google released the Generative AI report to a subset of UK sites first, and Google hasn't confirmed a timeline for the global rollout, so sites outside that first group may not see the report yet.

Until the Generative AI report reaches your Search Console property, use the impressions-up, clicks-flat pattern from the standard Performance report. Rising impressions paired with flat clicks on your target pages is the signal to watch in the meantime.

Engines use your content but never name you

Ghost citations are AI answers that use your content but never name your brand. ChatGPT, Perplexity, or Google AI Overviews may draw on your page to build an answer without naming your brand in the response, which leaves your work uncredited and invisible in brand mention tracking.

To reduce ghost citations, put your brand name and a clear, factual answer near the top of each section of your content. A section that opens with your brand name and a direct answer gives AI engines something attributable to quote. This guide on writing answer blocks AI engines quote walks through the format.

What to Do Next

Run the full loop for four weeks before you change anything. Then pull the 5 articles with the most impressions and the fewest citations. Those are your best candidates for a rewrite. Add short, sourced answer blocks under each heading and link to primary data. Then keep tracking brand citations in AI answer engines with the same prompt list, so before-and-after numbers stay comparable. If you want to cut the weekly manual work, read our guide on choosing an AI citation tracker.

Further reading

• SparkToro and Gumshoe.ai, AI recommendation consistency study (Jan 2026): explains why AI "rank positions" mislead.
• Seer Interactive, AI Overviews CTR research (Nov 2025): the baseline for how many clicks AI Overviews absorb.
• Seer Interactive, ghost citations work (Mar 2026): shows how your content gets used without your brand being named.
• Pew Research Center, AI Overviews browsing study (2025): an independent, non-vendor look at clicks across 900 U.S. adults.
• Ahrefs, AI visibility study of 75,000 brands (2026): shows which off-site mentions correlate with being cited.

Sources

1. Google AI Overviews drive 61% drop in organic CTR, 68% in paid — Search Engine Land (reporting Seer Interactive research), 2025. Supports: Organic CTR on informational queries fell 61% when an AI Overview appeared.
2. ChatGPT traffic analysis: Insights from 17 months of clickstream data — Semrush, 2026. Supports: ChatGPT's outbound referral traffic to the web grew 206% in 2025.
3. AI Recommendations Change With Nearly Every Query: SparkToro — Search Engine Journal, 2026. Supports: SparkToro and Gumshoe.ai found ChatGPT gave the same brand list twice less than 1 time in 100.
4. AI Citation Volatility: A 530,875-Citation Study — Mentions.ai, 2026. Supports: 69% of the sources behind a typical answer changed overnight, and 84% of cited sources were used by only one engine (vendor study).
5. LLM Ghost Citations: Why Your Content Is Working and Your Brand Isn't — Seer Interactive, 2026. Supports: Seer's March 2026 "ghost citations" research on content that gets used without the brand being named.
6. Across 75,000 Brands, YouTube Mentions Are the Strongest Signal of AI Visibility, New Ahrefs Report Reveals — Ahrefs via Business Wire, 2026. Supports: Ahrefs' 75,000-brand study found YouTube mentions had the strongest correlation with AI visibility.
7. How to Track ChatGPT, Gemini and Perplexity Referral Traffic in GA4 and CRM — DarwinApps, 2026. Supports: GA4's native AI Assistant channel, added in May 2026, leaves out Perplexity and sessions with no referrer.
8. Google Search Console Now Tracks AI Search: What to Do — Neil Patel, 2026. Supports: Google launched the Generative AI performance report on June 3, 2026, first to a subset of UK sites, and it reports impressions only.
9. Free AI Overviews Tracker by Ahrefs — Ahrefs, 2026. Supports: Ahrefs' December 2025 data showed AI Overviews cut clicks to top-ranking pages by about 58%.
10. Google users are less likely to click on links when an AI summary appears in the results — Pew Research Center, 2025. Supports: Pew's independent browsing study of how U.S. adults click when AI Overviews appear.
11. Ahrefs Brand Radar Review: Does It Meet Expectations? — Writesonic, 2026. Supports: Brand Radar reported 3 ChatGPT mentions for a brand that a manual check put at 123 (from a vendor that sells a competing tool).