AI assistants: an experiential agency case study
Anonymized Agency — AI Visibility & Analytics
AI-referred lead-form submissions increased 300%, from two to eight across comparable four-month periods. Sessions grew 186.9% over those same periods, giving this experiential agency a growing, measurable source of inquiries.
Small sample; form submissions have not been qualified. Comparison dates and tracking details are in the methodology.
Executive summary
More AI traffic, more recorded inquiries
An experiential marketing agency recorded 1,119 identifiable AI-assistant sessions, 797 distinct users, and 19 lead-form submission events across the full reporting year. Those events occurred in 19 converting sessions. AI was a measurable source of inquiries, accounting for 0.77% of website sessions and 1.65% of recorded forms.
Form submissions increased 300%.
Across comparable four-month periods after consistent form tracking began, submissions rose from two to eight. That is six additional recorded form events, with lead quality still unmeasured.
Traffic grew alongside inquiries.
AI sessions increased from 191 to 548 over those same four-month periods, a 186.9% increase. Forms per session rose from 1.05% to 1.46%.
Annual traffic increased 354.9%.
The full reporting year recorded 1,119 AI sessions, compared with 246 in the preceding year. AI’s share of website sessions increased from 0.36% to 0.77%.
Commercial pages produced most forms.
The homepage and three core service pages generated 16 of the 19 form events from 322 landing sessions. That is 84.2% of forms from about 28.8% of AI sessions.
One assistant generated every recorded form.
The leading assistant contributed 936 sessions and all 19 directly attributed forms. The other four measured assistants contributed 183 sessions without a recorded form.
The next step is measuring qualification.
Connect submissions to qualified opportunities and deal outcomes. The evidence supports developing the channel through its converting pages, while its commercial value remains unproven.
A growing inquiry channel, with a small sample. The growth comparison contains ten form events across both periods. The annual 19 events are not established as 19 unique qualified leads, additional sales caused by AI, or a revenue return. September data remains provisional as of the report date.
Company, client, platform, and other agency names have been removed, along with identifying business details and source-page URLs.
Traffic & submission growth
A comparable measure of growth
The opening growth figures use the same complete four-month windows after the form-tracking change. This allows the increase in submissions to be read alongside the increase in traffic, without comparing a tracked period with an untracked one.
| Measure | Earlier period | Later period | Change |
|---|---|---|---|
| AI sessions | 191 | 548 | +186.9% |
| Lead-form submissions | 2 | 8 | +300% |
| Forms per session | 1.05% | 1.46% | +0.41 percentage points |
The wider annual record also shows growing traffic. Full-year sessions increased 354.9%. Within the reporting year, the first complete quarter recorded 196 sessions and four forms; the last recorded 419 sessions and six forms. That is 113.8% more sessions and 50% more forms, while forms per session fell from 2.04% to 1.43%.
| Month | AI sessions | Forms | Session conversion rate |
|---|---|---|---|
| Oct | 81 | 2 | 2.47% |
| Nov | 72 | 2 | 2.78% |
| Dec | 43 | 0 | 0.00% |
| Jan | 75 | 2 | 2.67% |
| Feb | 63 | 1 | 1.59% |
| Mar | 74 | 1 | 1.35% |
| Apr | 55 | 2 | 3.64% |
| May | 107 | 1 | 0.93% |
| Jun | 129 | 2 | 1.55% |
| Jul | 95 | 2 | 2.11% |
| Aug | 159 | 3 | 1.89% |
| Sep | 165 | 1 | 0.61% |
The final month brought 165 sessions and one form, compared with 159 sessions and three forms in the month before. One additional submission would materially change the monthly rate. These small counts do not establish a persistent conversion decline or improvement.
Monthly session rows total 1,118; the full-period query returns 1,119. The annual headline uses the full-period result. The monthly rows have not been adjusted to force a match. Reporting windows and the growth calculations are documented in the methodology.
Conversions
An observed form rate comparable to paid search
AI’s annual session conversion rate was 1.70%, compared with 1.46% for paid search and 0.71% for organic search. These are observed differences between audiences and landing pages. A reliable performance advantage over paid search has not been established.
| Channel | Sessions | Form events | Session conversion rate | Engagement |
|---|---|---|---|---|
| AI assistants | 1,119 | 19 | 1.70% | 60.50% |
| Paid search | 58,076 | 852 | 1.46% | 65.27% |
| Organic search | 15,806 | 112 | 0.71% | 60.36% |
| Referral | 2,260 | 32 | 1.42% | 56.24% |
| Organic social | 4,308 | 15 | 0.35% | 55.57% |
| Direct | 59,956 | 104 | 0.17% | 32.47% |
| All website traffic | 144,756 | 1,154 | 0.79% | 50.30% |
Recognized AI sources are excluded from the comparison channels. Session conversion rate counts sessions containing a form event, rather than dividing every form event by sessions. Paid search had 848 converting sessions and 852 form events; the sitewide figures were 1,147 converting sessions and 1,154 events.
With 19 converting sessions, AI’s approximate 95% Wilson interval is 1.09%–2.64%, which includes the paid-search rate. That interval assumes independent sessions and does not account for repeat visitors or tracking errors.
AI visitors averaged 33.0 seconds of engagement per session, compared with 40.0 for paid search and 41.4 for organic search. Its 60.5% engagement rate was close to organic search’s and below paid search’s. The channel did not outperform every engagement measure.
- Desktop accounted for most inquiries. Desktop delivered 16 forms from 911 sessions. Mobile delivered three from 207 sessions, and tablet one session with no forms. These counts do not establish a device-specific conversion problem.
- Recorded forms were concentrated in the US. Sixteen of the 19 forms came from US-attributed sessions. Geography does not establish company location, buyer fit, or lead quality.
- First-touch attribution adds one associated event. Sixteen annual forms had the leading assistant as the user’s first recorded source. Fifteen overlapped with the 19 session-attributed forms, producing a union of 20 associated events. Adding 19 and 16 would double-count that overlap.
From answer to inquiry
Commercial pages do most of the conversion work
The homepage and the PR boxes, brand activation, and mobile tours service pages produced 16 of 19 form events from 322 landing sessions. These are the clearest places to improve the path from AI discovery to an inquiry.
| Landing-page group | Sessions | Forms | Forms per session |
|---|---|---|---|
| Blog | 294 | 0 | 0.00% |
| Homepage | 191 | 7 | 3.66% |
| Case studies | 191 | 2 | 1.05% |
| Unspecified landing page | 166 | 0 | 0.00% |
| Service pages | 132 | 9 | 6.82% |
| Other pages | 87 | 1 | 1.15% |
| Careers | 57 | 0 | 0.00% |
- PR boxes generated five forms from 79 sessions. Brand activation and mobile tours each generated two forms from 26 sessions. These small page-level samples indicate useful inquiry paths, rather than stable conversion benchmarks.
- Case studies also contributed. Two individual case-study pages each recorded one session and one form. Those one-for-one observations are too small to support a credible 100% conversion benchmark. The work portfolio generated one form from 51 sessions.
- Blog visits have not yet produced same-session forms. The 294 blog landing sessions had a 62.6% engagement rate and no recorded form events. This does not establish that the blog has no assisted value: later returns and research without a tracked conversion are outside this analysis.
- Some entry pages could not be identified. The 166 unspecified landing sessions contained no page views and only four engaged sessions. They remain in the report while collection is investigated. Removing them would raise the apparent form rate to about 1.99% without establishing better performance.
Improve the route from article to inquiry. Add relevant service links, project proof, and contextual inquiry prompts to articles already receiving AI traffic. Preserve the opportunity to measure later returns as well as same-session forms.
Landing-page rows total 1,118 sessions versus the 1,119 annual estimate. Landing pages are session entry pages, not necessarily the pages where forms were submitted.
AI visibility
Citations and converting visits tell different stories
The existing visibility snapshot covers 621 tracked prompts across four assistants. The two engines with the most linked citations cited the agency in 86 and 71 prompts respectively, or 13.85% and 11.43% of the prompt set. The assistant responsible for every directly attributed form had only four linked citations and six automated name matches.
Citations and converting visits therefore describe different outcomes. A citation rate cannot be treated as traffic share, conversion rate, or a sales forecast. The prompt sample mixes branded and nonbranded questions and does not represent real-world query volume.
- The prompt panel changed. The tracked set expanded from 461 to 621 prompts, and stored results mix probe dates. An apparent movement in an aggregate visibility score is not a controlled measure of improvement.
- Automated matches have limits. Individual responses were not all manually revalidated. The report reused the existing snapshot rather than running new probes.
- Coverage remains incomplete. Failed search-engine AI probes mean visibility is unmeasured, not zero. Search-engine AI impressions and their conversion contribution were not included in the referral total.
Unattributed app referrals, later direct visits, cross-device journeys, and research without a click may be missed. Source labels also do not prove that every visit came from a qualified human buyer. The totals measure recorded attribution, not the full economic impact of AI.
Measurement
Reliable growth starts with a consistent form count
Direct source reporting corrected the dashboard’s session, user, and conversion totals. A form-submission measure and a consistent historical definition are essential before interpreting growth.
| Measure | Existing dashboard | Verified report |
|---|---|---|
| AI sessions | 1,094 | 1,119 |
| AI conversions | 20 actions | 19 form events |
| AI users | 943 accumulated users | 797 distinct users |
| Prior-year AI sessions | 231 inferred from a stored change | 246 from the source query |
The session difference reflected missing daily records, incomplete assistant-domain coverage, and later revisions. The dashboard’s extra conversion was a call-to-action click. Its accumulated daily user values did not represent distinct users across the year. The report did not change the dashboard or its configuration.
Across all event names, AI sessions generated 271 key-event occurrences. Only 19 were the designated form-submission signal. Contact clicks, project-start clicks, and box-builder clicks were engagement measures and were excluded from form totals.
Count each submission once. An alternative form signal also recorded 19 events and may describe the same submissions. Combining the two signals into 38 leads would double-count an unresolved overlap. Only 11 form-start events were recorded, so these aggregates cannot establish a valid form-completion funnel.
Recommendations
Build on the pages already producing inquiries
Prioritize measurable, qualified inquiries. The following actions are recommendations from the analysis; their effect has not yet been measured.
1. Repair the measurement baseline.
In the first two weeks, align conversions to the designated form event, repair missing source history, complete assistant-domain coverage, and investigate unspecified landing pages and form-start collection. Map each form to one server-side submission ID.
2. Improve the converting pages.
Review the homepage and three core service pages for clear buyer fit, capabilities, project evidence, inquiry expectations, and an accessible form. These four entry pages produced 16 of 19 forms.
3. Connect articles to services.
Add relevant service links, case studies, and contextual inquiry prompts to articles with demonstrated AI traffic. Measure downstream service visits and eventual deduplicated inquiries.
4. Join submissions to qualification.
Connect form IDs to CRM lifecycle stages while retaining first source, session source, landing page, and submitted service interest. Track forms through qualified leads, opportunities, and won revenue.
5. Stabilize visibility measurement.
Separate branded and nonbranded prompts, retain a consistent prompt panel, and distinguish successful answers from errors. Keep citation measures separate from traffic and conversion outcomes.
6. Review investment after more evidence.
Continue a bounded program within the existing plan, then review conversion volume, lead quality, and production cost together. Nineteen annual form events do not justify a major budget shift on conversion-rate evidence alone.
Use a monthly scorecard with explicit stages. Report sessions, distinct users, form-producing sessions, deduplicated submissions, qualified leads, opportunities, won revenue, and allocated acquisition cost. Keep citations and call-to-action events as supporting measures.
Method & caveats
The comparison behind the growth figures
The supplied analytics report used read-only source queries verified on October 1, 2026. The full reporting window is October 1, 2025–September 30, 2026; the prior traffic window is October 1, 2024–September 30, 2025. Queries use the property’s reporting timezone.
The opening growth comparison uses June–September 2026 versus June–September 2025, complete four-month windows after the primary form event began being tracked in May 2025. Submissions increased from two to eight: (8 − 2) ÷ 2 × 100 = 300%. Monthly session totals increased from 191 to 548: (548 − 191) ÷ 191 × 100 = 186.9%, rounded.
A full-year form-growth comparison is not valid because most of the earlier year lacks the same conversion definition. Annual traffic growth is a separate comparison: (1,119 − 246) ÷ 246 × 100 = 354.9%. The first-to-last-quarter comparison uses October–December 2025 and July–September 2026.
How the figures were counted
- AI sessions. Known assistant referral hosts and exact source names were matched regardless of medium. Lookalike or unverified domains were excluded. Search-engine organic traffic was not relabeled as AI traffic.
- Form conversions. The primary form event was checked against key events. All 19 annual events reconcile across month, source, landing page, device, and country reports. Session conversion rate counts sessions containing the event.
- Reporting reconciliation. Provider and device totals reconcile to 1,119 sessions. Monthly and landing-page totals are 1,118; daily totals are 1,120. The annual headline uses the full-period estimate. The source report disclosed these small aggregation differences.
- Visibility snapshot. The report read existing successful probes spanning September 9–17, 2026. Automated matching, a changing prompt panel, mixed probe dates, and incomplete engine coverage limit comparisons.
Limits to the findings
- Small samples. The matched form-growth comparison contains two events in the earlier period and eight in the later one. These ten observations do not prove better qualification, a lasting conversion increase, or causal lift.
- Provisional close. September is a complete calendar month but remains provisional as of October 1. Recheck the final days after processing before treating the month as a locked close.
- Unverified commercial outcomes. Unique submissions after server-side deduplication, qualified leads, opportunities, sales, acquisition costs, and ROI were not established. No recorded purchase revenue in analytics does not prove that associated buyers generated no revenue.
- Incomplete attribution. Repeat visitors, untracked research, stripped referrers, and cross-device paths limit what can be inferred. There was no controlled test or complete multi-touch journey export establishing that optimization caused the observed growth.