ChatGPT Search Referral: Cara Membaca utm_source=chatgpt.com di Analytics

Suatu pagi, analyst membuka acquisition report dan melihat source baru yang mulai konsisten.

chatgpt.com.

Jumlahnya belum besar.

Tetapi conversion rate lumayan.

Marketing Lead langsung bertanya:

“Ini berarti semua traffic dari ChatGPT Search?”

Jawabannya perlu hati-hati.

OpenAI menjelaskan bahwa referral URLs dari ChatGPT Search secara otomatis menyertakan parameter UTM `utm_source=chatgpt.com`. Ini membuat inbound traffic yang datang melalui link dari ChatGPT lebih mudah dilacak di analytics platform.

Itu kabar bagus.

Tetapi analytics tetap harus dibaca sesuai scope.

UTM memberi signal referral click.

Ia tidak mengukur semua exposure di ChatGPT.

Brand bisa disebut tanpa click.

User bisa melihat source lalu datang direct.

User bisa copy URL.

Journey bisa berlanjut di device lain.

Jadi `utm_source=chatgpt.com` adalah attribution signal yang kuat untuk click-through yang teridentifikasi, bukan census seluruh AI visibility.

Mulai dari definisi

Ketika user mengklik link dari ChatGPT dan referral URL membawa parameter:

`utm_source=chatgpt.com`

analytics dapat menangkap source tersebut.

Di Google Analytics atau platform lain, hasil parsing bisa terlihat sebagai:

source = chatgpt.com.

Medium tergantung konfigurasi.

Campaign fields mungkin kosong jika tidak ada parameter lain.

Yang penting adalah source.

OpenAI sengaja menambahkan UTM agar publisher bisa melacak inbound traffic.

Jadi jika property Anda mengizinkan OAI-SearchBot dan content muncul di Search results ChatGPT, clicks bisa lebih mudah diidentifikasi.

Ini lebih baik dibanding traffic yang datang tanpa referrer atau tanpa campaign parameter.

Tapi jangan mengandalkan default channel grouping tanpa audit.

GA4 bisa mengelompokkan dengan cara tertentu.

Custom analytics bisa beda.

Pastikan raw source terlihat.

Jangan hanya lihat “Referral” total.

Buat dimension source.

Filter `chatgpt.com`.

Apa yang harus dicek pertama?

Volume.

Users.

Sessions.

Landing pages.

Engagement.

Conversions.

Revenue atau lead.

New vs returning.

Device.

Country.

Date trend.

Kemudian bandingkan dengan channel lain.

Tidak perlu langsung membuat dashboard kompleks.

Pertanyaan basic:

Traffic masuk ke page apa?

Apakah page tersebut sesuai query intent?

Apakah user melakukan sesuatu?

Jika 80% ChatGPT traffic landing di article teknis dan 5% masuk service page, behavior-nya mungkin early research.

Jika landing di comparison atau pricing, intent lebih commercial.

Landing page memberi context.

Referral bukan visibility

Ini distinction utama.

Misalnya ChatGPT menyebut brand 1.000 kali.

Hanya 50 user click.

Analytics melihat 50.

Bukan 1.000.

Atau ChatGPT menampilkan link 1.000 kali, 100 click.

Analytics melihat clicks yang berhasil sampai dan di-track.

Jadi jangan membuat statement:

“ChatGPT visibility = 100 sessions.”

Sessions adalah visit.

Visibility berada upstream.

Kalau ingin melihat visibility, perlu controlled observation, brand mention testing, source tracking, atau platform data lain jika tersedia.

Referral analytics adalah after-click measurement.

Ini mirip Search.

Search Console impressions dan Analytics sessions tidak sama.

Untuk ChatGPT, gap bisa lebih besar karena answer sering memberi information directly.

High conversion rate bisa misleading kalau volume kecil

10 sessions.

2 leads.

Conversion rate 20%.

Marketing senang.

Tetapi sample sangat kecil.

Minggu berikutnya 30 sessions, 1 lead.

Rate turun.

Jangan membuat conclusion dari base kecil.

Track absolute.

Use rolling period.

Monthly.

Quarterly.

B2B dengan low volume mungkin perlu 3-month view.

Publisher high volume bisa weekly.

Statistics basic still applies.

AI referral bukan exception.

Segmentation matters.

Landing page type lebih berguna daripada total

Group:

Blog.

Research.

Service.

Pricing.

Case Study.

Contact.

Product.

Local.

Then see.

Maybe ChatGPT traffic over-indexes on research pages.

Good signal that source content is being surfaced.

Maybe service page traffic converts strongly.

That is commercially valuable.

Maybe old article gets traffic but info outdated.

Risk.

Analytics can become maintenance trigger.

If ChatGPT sends users to old page, review freshness.

This is practical AI Optimization.

Don't only chase more sessions.

Ensure destination still correct.

Query information is not automatically available

UTM source tells you origin.

It does not tell you exact prompt the user typed.

Do not invent.

A visitor arrives from ChatGPT to article “OAI-SearchBot Guide.”

You can infer likely topic.

But not exact prompt.

Don't write CRM field:

“Prompt: how to configure robots.txt”

unless user explicitly provided it.

Source attribution and query attribution are different.

This is important for privacy and methodology.

Controlled query testing can give hypotheses.

Analytics gives actual visits.

But don't join one manual query to one anonymous session without evidence.

Referrer and UTM can be lost

Analytics isn't perfect.

Browser privacy.

Redirect.

Consent mode.

Ad blockers.

Cross-domain setup.

Server-side redirects.

URL cleaning.

Link shorteners.

Some clicks may arrive without full parameters.

So source count may underrepresent.

Test.

Click from ChatGPT.

Inspect landing URL.

GA realtime.

Network.

Session source.

Make sure WordPress plugin or redirect doesn't strip UTM.

Some sites aggressively clean query parameters for canonical reasons.

Cleaning visible URL after analytics capture is fine if implemented correctly.

Stripping before analytics sees parameter loses attribution.

SEO canonical should usually ignore UTM parameters and point to clean canonical URL.

But don't redirect every UTM URL to clean URL in a way that destroys campaign data before capture.

Measurement and canonicalization need coordination.

UTM parameter should not create duplicate index problems

A URL like:

/article/?utm_source=chatgpt.com

should generally resolve as same content.

Canonical points to:

/article/

Search engines understand tracking parameters in context.

Don't create separate page based on UTM.

Don't add UTM URL to sitemap.

Analytics param is attribution.

Not content identity.

If caching system treats every UTM as new variant, optimize cache config.

Don't let analytics parameter create performance issue.

ChatGPT referral should be its own reporting row

In management dashboard, don't hide inside “Referral.”

Create row:

ChatGPT identifiable referral.

Sessions.

Users.

Engaged sessions.

Conversions.

Conversion rate.

Top landing pages.

Revenue/pipeline if applicable.

Then compare:

Google Organic.

Direct.

LinkedIn.

ChatGPT.

Other AI sources.

Keep naming precise.

“Identifiable ChatGPT referral.”

Why “identifiable”?

Because not all AI-assisted journeys are captured.

This prevents stakeholder from treating number as total ChatGPT impact.

Avoid labels like:

“ChatGPT traffic total.”

Too absolute.

B2B should push data into CRM

If user fills demo form, capture source.

UTM can persist into form hidden field.

But implement carefully.

Store original source and latest source separately if possible.

Why?

Buyer may first come from ChatGPT, leave, return via Google.

If form only stores last-touch, ChatGPT influence disappears.

For B2B long cycle, useful fields:

first_touch_source.

last_touch_source.

landing_page.

conversion_page.

self_reported_discovery.

Then sales can ask:

“How did you first hear about us?”

Answer options include:

ChatGPT or AI assistant.

Google Search.

Referral.

Event.

Social.

Colleague.

Other.

Self-reported data complements technical attribution.

No need pretend perfect.

Publisher should look beyond sessions

For publisher, metrics:

pages/session.

engaged time.

newsletter signup.

subscription.

return visit.

article category.

If ChatGPT referrals go to deep research pages and readers stay longer, valuable.

If they bounce after five seconds, maybe answer mismatch.

Maybe ChatGPT summary already gave enough.

Maybe page slow.

Maybe paywall.

Investigate.

No universal benchmark.

Compare to site baseline.

E-commerce should trace product behavior

ChatGPT referral to product page.

Then:

view item.

select variant.

add to cart.

checkout.

purchase.

return.

Product fit matters.

High referral but high return could mean AI recommendation mismatch or product page ambiguity.

Don't optimize only click.

Commerce outcome is purchase quality.

If ChatGPT sends 100 visitors and 20 buy but 15 return, not great.

Attribution must connect to commerce data.

Referral spike can come from one citation

Sometimes article gets cited in a popular conversation pattern.

Traffic spikes.

Don't assume broad visibility.

Check landing concentration.

If 90% visits hit one URL, it may be one topic.

Create content cluster if strategic.

But don't mass produce.

Could be temporary news cycle.

Use trend.

If referral remains across many pages, stronger sign of footprint expansion.

Source diversity is interesting.

Maybe 50 pages each get small traffic.

That indicates broader discovery.

Can we measure ChatGPT Search CTR?

Not from analytics alone.

CTR needs denominator:

impressions or link exposures.

Analytics gives clicks/sessions.

OpenAI does not provide publisher dashboard with total link impressions in this context.

So do not calculate:

ChatGPT CTR = sessions / manual prompt count.

Denominator mismatched.

Third-party tools can estimate visibility.

But label methodology.

Referral analytics cannot give true Search CTR without exposure data.

This is a major limitation.

Can we compare conversion with Google?

Yes, but control for intent.

Google Organic includes many query types.

ChatGPT traffic may skew research-heavy.

Compare landing page categories.

Same funnel stage.

For example:

ChatGPT service-page sessions vs Google service-page sessions.

More fair.

Even then, attribution differs.

Use directional insight.

Not universal claim:

“ChatGPT traffic converts 3x better.”

Unless sample large and methodology sound.

OpenAI UTM is useful precisely because it makes direct click easier to see

This point shouldn't be lost.

Many AI journeys are hard to measure.

The automatic `utm_source=chatgpt.com` is a valuable practical feature.

It lets publishers see:

what pages ChatGPT sends traffic to.

which conversions follow.

how traffic changes over time.

This should be used.

Don't ignore because attribution imperfect.

Measurement doesn't need perfection to be useful.

It needs known limitations.

Build monthly ChatGPT referral review

Simple:

Sessions.

Users.

Top landing pages.

Conversions.

Conversion value.

Country.

New vs returning.

Trend.

Anomalies.

Then connect to:

AI query observations.

content updates.

PR.

site changes.

Maybe article update leads to more referral.

Maybe.

Record correlation.

Don't claim causation automatically.

Referral data is one layer.

Change log is another.

Do not remove UTM because URL looks ugly

Some brand teams dislike query strings.

“Kenapa link-nya ada utm_source?”

Because measurement.

It's okay.

Canonical URL remains clean.

Browser URL can include parameter.

If you want to clean after page load, do so after analytics captures and without breaking session attribution.

Test thoroughly.

Don't create redirect chains.

Performance matters.

Attribution should not hurt UX.

What if ChatGPT traffic appears as direct?

Possible if UTM is stripped or user copy-pastes.

Audit.

Test several scenarios.

Open ChatGPT result.

Click.

See URL.

Check GA realtime.

If source missing:

consent?

redirect?

tracking code?

cross-domain?

server-side?

Don't assume OpenAI not sending UTM.

Official guidance says it automatically includes the parameter in referral URLs.

If your stack loses it, troubleshoot stack.

At the end, what does `utm_source=chatgpt.com` tell us?

It tells us something valuable and limited:

a user clicked through from a ChatGPT referral URL that carried the parameter and reached a trackable page.

It does not tell us:

how many people saw brand.

how many answers mentioned site.

what exact prompt.

what ranking.

how many no-click exposures.

whether ChatGPT caused later direct visits.

That's okay.

One metric doesn't need to answer everything.

The mistake isn't incomplete data.

The mistake is pretending incomplete data is complete.

For AI Optimization, `utm_source=chatgpt.com` should sit in the outcome layer.

Upstream:

crawlability.

Search inclusion.

citation.

brand mention.

Downstream:

engagement.

lead.

revenue.

retention.

When these layers are connected, ChatGPT Search stops being a screenshot topic and becomes a measurable acquisition channel, at least for the portion of journeys that actually click through.

That's already much more useful than guessing.