2026-10-30 21:36:51
Between March 2025 and September 2026, our research team tracked client retention across 418 U.S. marketing agencies, covering roughly 19,700 client relationships. Participating agencies shared anonymized billing records, which we used to log the month each client started and the month each relationship ended.
Customer retention rate (CRR) in this report is the share of an agency’s clients at the start of a month who are still paying at the end of it, excluding clients added during that month. We measured CRR monthly because most agency engagements, including retainers, are billed and cancelled month to month. A monthly figure also converts directly into client lifetime: average lifetime in months equals 1 divided by (1 minus monthly CRR). An agency that keeps 90.9% of its clients each month holds the average client for 11 months.
Agencies were grouped by the service that made up the largest share of their revenue. We normalized the dataset by excluding one-time projects under $5,000 and clients lost through acquisition or bankruptcy, so the figures reflect clients who chose to leave. In the sections below, we break down CRR and client lifetime by agency type, agency size, and client company size, followed by the reasons clients gave for ending the relationship.
Across all 14 agency types in our dataset, the average client stays for 12.3 months. The spread between types is wide: advertising agencies keep the average client for more than two years, while web design firms keep them for under seven months. In the table below, we break down monthly CRR, the share of clients still active after 12 months, and average client lifetime for each agency type.
Average Customer Retention Rate and Client Lifetime by Agency Type (September 2026)
| Agency Type | Average Monthly CRR | Clients Retained After 12 Months | Average Client Lifetime |
| Advertising (Creative and Media) | 96.4% | 64.4% | 27.8 months |
| Full-Service Digital Marketing | 95.2% | 55.4% | 20.8 months |
| Public Relations | 93.6% | 45.2% | 15.6 months |
| Email and Marketing Automation | 92.7% | 40.3% | 13.7 months |
| Content Marketing | 91.8% | 35.8% | 12.2 months |
| GEO and AI Search | 91.4% | 34.0% | 11.6 months |
| SEO | 90.9% | 31.8% | 11.0 months |
| Marketing Strategy and Consulting | 90.3% | 29.4% | 10.3 months |
| Branding and Creative | 89.6% | 26.8% | 9.6 months |
| Lead Generation | 89.1% | 25.0% | 9.2 months |
| PPC and Paid Media | 88.3% | 22.5% | 8.5 months |
| Social Media Marketing | 87.6% | 20.4% | 8.1 months |
| Influencer Marketing | 86.2% | 16.8% | 7.2 months |
| Web Design and Development | 85.1% | 14.4% | 6.7 months |
Our researchers took three main points from this data:
Average lifetime tells only part of the story, because clients do not leave at an even pace. To see when departures happen, we followed every client relationship from its first billed month and recorded the share still active at each milestone. The table below compares four agency types from different points on the retention range.
Share of Clients Still Active by Month of Engagement (September 2026)
| Month of Engagement | SEO | Full-Service Digital Marketing | PPC and Paid Media | Web Design and Development |
| Month 1 | 96.2% | 98.1% | 94.8% | 97.3% |
| Month 3 | 81.7% | 91.4% | 76.3% | 83.9% |
| Month 6 | 52.4% | 77.2% | 46.9% | 49.6% |
| Month 9 | 39.8% | 67.5% | 32.4% | 28.1% |
| Month 12 | 32.6% | 57.1% | 23.7% | 15.2% |
| Month 18 | 20.3% | 41.8% | 12.6% | 7.4% |
| Month 24 | 12.9% | 31.6% | 6.8% | 3.9% |
Our team drew three conclusions from the survival data:
Agency size had a strong effect on retention in our dataset. We grouped agencies by full-time headcount and calculated monthly CRR and average client lifetime for each group, across all service types.
Average Customer Retention Rate by Agency Size (September 2026)
| Agency Size (Employees) | Average Monthly CRR | Clients Retained After 12 Months | Average Client Lifetime |
| 1 to 10 | 88.4% | 22.8% | 8.6 months |
| 11 to 50 | 90.6% | 30.6% | 10.6 months |
| 51 to 200 | 92.9% | 41.3% | 14.1 months |
| More than 200 | 94.8% | 52.7% | 19.2 months |
Three findings stood out to our team:
The size of the client mattered as much as the size of the agency. We grouped each client relationship by the client’s annual revenue and calculated retention across all agency types.
Average Customer Retention Rate by Client Company Size (September 2026)
| Client Company Size (Annual Revenue) | Average Monthly CRR | Clients Retained After 12 Months | Average Client Lifetime |
| Small Business (Under $10M) | 87.9% | 21.3% | 8.3 months |
| Lower Mid-Market ($10M to $100M) | 90.8% | 31.4% | 10.9 months |
| Upper Mid-Market ($100M to $1B) | 93.3% | 43.5% | 14.9 months |
| Enterprise (Over $1B) | 95.6% | 58.3% | 22.7 months |
Our researchers noted three patterns in this breakdown:
When a relationship ended, participating agencies recorded the main reason the client gave, either in the cancellation notice or in an exit conversation. We grouped those reasons into seven categories and calculated each one’s share of all departures.
Our researchers drew three conclusions from the exit data:
If you’d like to request a PDF copy of this report or learn more about our agency, you can reach out here.
2026-10-29 23:40:16
From September 8 to October 25, 2026, our research team studied Muse usage across the United States and Canada, the two markets where it was available during that period. We combined aggregated app-intelligence data with activity logs from a panel of 4,180 Muse users who opted into our research, then weighted the panel to match the age and regional mix of the full user base.
Meta launched Muse, its personal AI agent, on September 8, 2026. Muse connects to a user’s email, calendar, bank and shopping accounts, then carries out tasks such as sending messages, booking reservations and completing purchases once the user approves them. Because Muse acts on a person’s behalf, the usual chatbot measures of prompts and sessions tell only part of the story, so our study also tracked what users hand off to it and how often they return. For broader context on this category, see our agentic AI statistics.
In the sections below, we break down Muse’s overall usage, its weekly growth, the activities people use it for, its users by age group, and how its retention compares with other AI assistants.
In the table below, we break down the core usage metrics our team recorded for Muse during its first seven weeks on the market.
Meta Muse Usage Statistics (September-October 2026)
| Metric | Value |
| Accounts set up in the U.S. and Canada (as of October 25) | 11.2 million |
| Average daily active users (week ending October 25) | 2.74 million |
| Average weekly active users (week ending October 25) | 4.96 million |
| Daily-use rate (daily active users as a share of accounts) | 24.5% |
| Weekly active users who opened Muse daily | 55.2% |
| Day 1 retention | 71.8% |
| Day 30 retention | 16.4% |
| Average sessions per daily active user | 3.4 |
| Average time spent per daily active user each day | 14.7 minutes |
| Sessions started from a suggestion by Muse | 27.4% |
| Weekly active users with at least one connected account | 58.3% |
| Weekly active users who completed a purchase through Muse | 22.8% |
| Average order value of purchases made through Muse | $174 |
| Accounts on a paid plan (Power or Maximum) | 3.1% |
| Users aged 45 and older | 52.2% |
Our team took three findings from this data:
In the table below, we track Muse’s cumulative accounts and average daily active users for each week of the study, with weeks ending on Sunday. The first week covers September 8 to 13.
Meta Muse Weekly Growth (September-October 2026)
| Week Ending | New Accounts | Cumulative Accounts | Average Daily Active Users | Daily-Use Rate |
| September 13 | 1.92 million | 1.92 million | 0.84 million | 43.8% |
| September 20 | 1.39 million | 3.31 million | 1.12 million | 33.8% |
| September 27 | 1.55 million | 4.86 million | 1.41 million | 29.0% |
| October 4 | 1.26 million | 6.12 million | 1.63 million | 26.6% |
| October 11 | 1.56 million | 7.68 million | 1.97 million | 25.7% |
| October 18 | 1.79 million | 9.47 million | 2.38 million | 25.1% |
| October 25 | 1.73 million | 11.20 million | 2.74 million | 24.5% |
Our researchers drew three conclusions from the weekly data:
In the table below, we break down what people ask Muse to do. The first column shows the share of weekly active users who used Muse for each activity at least once in a week, and the second shows each activity’s share of all tasks users handed to Muse.
Meta Muse Usage by Activity (September-October 2026)
| Activity | Share of Weekly Active Users | Share of All Tasks |
| Shopping and product research | 47.2% | 24.6% |
| Email and inbox management | 38.9% | 17.3% |
| Calendar and scheduling | 33.5% | 12.1% |
| Writing and messaging help | 29.8% | 11.4% |
| Companionship and conversation | 17.1% | 9.6% |
| Personal finance and budgeting | 21.7% | 7.8% |
| Travel and dining plans | 19.4% | 6.9% |
| Health and fitness tracking | 14.6% | 5.2% |
| Small business tasks | 6.3% | 3.1% |
| Other activities | 8.4% | 2.0% |
Our team took three findings from the activity data:
In the table below, we break down Muse’s user base by age group, along with how often each group uses Muse and what it uses Muse for most. Muse is available only to adults 18 and older.
Meta Muse Users by Age Group (September-October 2026)
| Age Group | Share of Muse Users | Daily-Use Rate | Average Tasks per Active Week | Most Common Activity |
| 18-24 | 8.9% | 19.8% | 9.2 | Writing and messaging help |
| 25-34 | 17.6% | 22.4% | 12.7 | Shopping and product research |
| 35-44 | 21.3% | 24.1% | 14.9 | Email and inbox management |
| 45-54 | 22.8% | 25.9% | 15.6 | Shopping and product research |
| 55-64 | 17.2% | 26.1% | 14.1 | Shopping and product research |
| 65 and older | 12.2% | 26.4% | 12.8 | Companionship and conversation |
Our researchers drew three conclusions from the age data:
In the table below, we compare Muse’s retention with that of other leading AI assistants. Each figure is the share of U.S. users who installed the app in September 2026 and returned to it on the given day after install. For usage data on the other platforms, see our ChatGPT usage statistics and our ranking of the top generative AI chatbots.
AI Assistant Retention by Days Since Install (September-October 2026)
| AI Assistant | Day 1 | Day 7 | Day 14 | Day 30 |
| Meta Muse | 71.8% | 28.6% | 21.9% | 16.4% |
| ChatGPT | 39.4% | 24.2% | 20.6% | 17.8% |
| Claude | 34.1% | 18.3% | 15.2% | 12.7% |
| Meta AI | 31.8% | 15.6% | 12.4% | 10.1% |
| Grok | 29.6% | 14.8% | 11.9% | 9.4% |
Our team took three findings from the retention data:
If you would like to request a PDF copy of this report or learn more about our agency, you can reach out here.
2026-10-29 23:27:08
Agentic commerce moved from concept to mass-market in the fall of 2026. Meta’s Muse agent reached more than 10 million accounts within seven weeks of its September 8 launch, and shopping became its most common use. Shopify opened Shop Pay checkout to Muse across its merchants, while Amazon blocked Muse from its store. ChatGPT, Claude and Google Gemini (including AI Mode in Google Search) all offer their own shopping and checkout features, and two open standards, the Agentic Commerce Protocol (ACP) and the Universal Commerce Protocol (UCP), now define how agents read catalogs and place orders.
For an ecommerce brand, this creates a new type of visitor. An AI agent reads your product data, checks your prices and policies, decides whether you are the right store, and then tries to buy. If any step fails, the agent moves on to a competitor, and the shopper never sees your site at all. In this guide, we explain how agents shop, where they fail on today’s ecommerce sites, and the work required to make your store ready, based on our Agentic Search Optimization work with ecommerce clients.
Every agent purchase moves through the same three stages that our Agentic Search Optimization framework describes: Retrieval, Evaluation and Action. Rankings are decided at Retrieval, since an agent can only choose among the stores on its shortlist. In ordinary AI search, a human does the Evaluation; in agentic commerce, the agent evaluates the options itself and only asks the shopper to approve its pick.
Underneath all three stages is verification. An agent cross-checks what your product pages say against your feed, your checkout and what other sites say about you. When the price on your page differs from the price in your feed, or a third-party review lists a different return window, the agent treats your store as a risk and moves on. Consistent information across your site and a diverse set of third-party sources matters more in agentic search than in any channel before it.
Between August and October 2026, our team ran 2,400 test purchases with AI shopping agents across 160 ecommerce sites in apparel, home goods, electronics and beauty. In 38.6% of attempts, the agent failed to complete the order. The chart below breaks down those failures by cause.
Our team took three lessons from the testing:
An agent-ready store has four technical layers between the agent and a completed sale, plus a set of trust signals that live on other websites. The diagram below shows how they fit together. Each layer removes one of the failure causes from the previous section.
The four on-site layers decide whether an agent can buy from you. The off-site trust signals decide whether it wants to. A store that is technically perfect but absent from reviews and best-of lists will still lose the Evaluation stage to a competitor that agents see recommended across many sources.
The work below follows the four layers of the stack, then the off-site signals. Most stores can complete the first two steps within a month.
Start with your bot management settings, since a blocked agent never sees anything else you have built. Most providers, including Cloudflare, Akamai and Fastly, now let you allow verified AI agents as a separate category from scrapers. Review the rules for your product, cart and checkout pages, and confirm that rate limits allow an agent to load several product pages and a checkout in under a minute.
Decide on a policy for each agent rather than blocking all of them. Amazon chose to block Muse and keep shoppers inside its own app, which works for a marketplace with its own traffic. Most brands do better letting agents buy, because an agent that cannot buy from you buys the same item from a competitor.
Agents compare stores on facts, so every fact needs to be present, current and identical everywhere it appears. Your product feed, product pages, structured data and checkout should all show the same price, stock level, variants and delivery estimate. Ecommerce is the one area where we consider structured data essential, because agents use it to confirm what the page says. A minimal example for one product looks like this:
{
“@context”: “https://schema.org”,
“@type”: “Product”,
“name”: “Ridgeline Waterproof Hiking Jacket”,
“sku”: “RWJ-2047-BLK-M”,
“gtin13”: “0812345020473”,
“color”: “Black”,
“size”: “M”,
“offers”: {
“@type”: “Offer”,
“price”: “229.00”,
“priceCurrency”: “USD”,
“availability”: “https://schema.org/InStock”,
“shippingDetails”: {
“@type”: “OfferShippingDetails”,
“deliveryTime”: {
“@type”: “ShippingDeliveryTime”,
“transitTime”: { “@type”: “QuantitativeValue”, “minValue”: 2, “maxValue”: 3, “unitCode”: “DAY” }
}
},
“hasMerchantReturnPolicy”: {
“@type”: “MerchantReturnPolicy”,
“merchantReturnDays”: 30,
“returnFees”: “https://schema.org/FreeReturn”
}
}
}
Write your shipping, returns and warranty policies in plain sentences with exact numbers, such as “free returns within 30 days,” so an agent can quote them to the shopper without guessing.
The Universal Commerce Protocol (UCP), led by Google and Shopify, lets agents read your catalog and place orders through a standard manifest. The Agentic Commerce Protocol (ACP), built by OpenAI and Stripe, powers checkout inside ChatGPT. Shopify, Wix and other platforms are adding these protocols for their merchants, so check what your platform supports before building anything custom. Stores on custom platforms can add a gateway that translates both protocols into their existing order system.
The diagram below shows a typical protocol checkout. The agent never handles the shopper’s card number; it passes a one-time payment token that your store charges like any other order.
Walk through your own checkout and note every step that assumes a person is there. Forced account creation, CAPTCHAs, newsletter pop-ups, chat widgets that cover the buy button and shipping costs revealed only on the last page all stop agents. Offer guest checkout and at least one tokenized wallet that agents support, such as Shop Pay, Stripe Link or PayPal.
Once an agent can buy from you, it still has to choose you. Agents favor stores that many independent sources describe as the best option for the shopper’s specific need. That means reviews on retail and review sites, placements on best-of lists that rank on Google, and press coverage that repeats the same claims about your brand. Pages on your own site for each type of shopper and use case, such as “waterproof jackets for long-distance hikers,” help an agent match you to the right requests.
| Area | What to Check | How to Test It |
| Access | Verified AI agents are allowed on product, cart and checkout pages | Review bot management logs for blocked agent requests |
| Access | Rate limits allow a full purchase in under a minute | Run a scripted purchase and watch for throttling |
| Product data | Price, stock and variants match across feed, page and checkout | Compare 50 random products across all three sources |
| Product data | Every product has a GTIN, clear variant names and a delivery estimate | Audit the feed for empty fields |
| Product data | Shipping, return and warranty terms state exact numbers | Read each policy page and highlight every vague phrase |
| Protocols | Your platform’s UCP and ACP support is turned on | Check your platform’s agentic commerce settings |
| Checkout | Guest checkout works with no account, CAPTCHA or pop-up | Complete a purchase in a fresh browser with no cookies |
| Checkout | At least one agent-supported wallet is offered | Test checkout with Shop Pay, Stripe Link or PayPal |
| Off-site trust | Your store appears in reviews and best-of lists for your main products | Ask ChatGPT, Claude and Google Gemini to recommend a store for those products |
| Off-site trust | Prices and policies match on every third-party site | Compare your top marketplace and review listings with your site |
Not every task carries the same weight. The matrix below places each task by its effect on agent sales and the effort it takes, based on our work with ecommerce clients.
The four quick wins remove the most common failure causes in a few weeks and usually need no new development. Product data cleanup and off-site reviews and list placements take longer, but they decide whether an agent chooses your store once it can buy from you, so they belong in the plan from the first month. A full checkout redesign or a custom MCP server rarely pays off before the basics are in place.
We score ecommerce sites on four levels of agent readiness. Each level depends on the one below it, so a store cannot be recommended by agents until it is transactable.
Of the 160 sites in our testing, 30.6% were at Level 1, 43.8% at Level 2, 21.9% at Level 3 and only 3.8% at Level 4. The step from Level 2 to Level 3 is mostly technical and can often be made within a quarter. The step to Level 4 depends on off-site reputation, which takes longer to build and is much harder for competitors to copy.
| Level | What Agents Can Do | Main Work to Reach the Next Level |
| 1. Blocked | Nothing; requests are refused or data is missing | Allow verified agents and fill gaps in the product feed |
| 2. Readable | Read products, prices and policies | Turn on protocol support and remove human-only checkout steps |
| 3. Transactable | Complete a purchase | Earn reviews and best-of list placements and keep facts consistent everywhere |
| 4. Recommended | Choose your store first for your core products | Maintain data accuracy and expand into new product categories |
Share this checklist with your developers and ecommerce team to get started. The table that follows links to the official developer documentation for each area.
| Resource | Publisher | What It Covers |
| Universal Commerce Protocol (UCP) integration guides | Google for Developers | Building and testing checkout and cart endpoints for UCP-powered checkout on Google |
| About UCP-powered checkout on Google | Google Merchant Center Help | Eligibility and setup steps for merchants |
| Intro to Product Structured Data | Google Search Central | Product, Offer, shipping and return markup for product pages |
| Product Feed Reference | OpenAI Developers | Product feed format for appearing and selling in ChatGPT |
| Agentic Commerce | Stripe Docs | Selling to agents, catalog feeds and shared payment tokens |
| Agentic Commerce | Shopify Developers | How agents search, build carts and check out on Shopify stores |
| Verified Bots | Cloudflare Docs | Identifying and allowing trustworthy bots and AI agents |
| Schema.org Product | Schema.org | The full list of product properties |
Shoppers buy where they research, and a growing share of research now happens inside AI agents. As the platforms add ads and paid checkout features to AI search, the stores that agents already recommend organically will keep winning sales without paying for them. The work to become one of those stores starts with the technical basics in this guide and continues with the off-site reputation that leads agents to choose you.
First Page Sage helps ecommerce brands become the store that AI platforms and agents recommend. If you would like an agentic commerce readiness audit of your site, you can reach out here.
2026-10-29 04:05:38
Last updated October 28, 2026
The following report was compiled from data collected between May 2025 and October 2026 from approximately 170 client companies running search and AI visibility campaigns with our agency. For each company, our research team isolated the unique visitors referred specifically by ChatGPT and measured the percentage of those visitors who completed an action the company counts as a “conversion.”
Most companies in the dataset invested in generative engine optimization (GEO) and, as a direct result, saw higher traffic specifically from ChatGPT and built conversion funnels tailored to that channel. As more users search through ChatGPT and the platform expands its paid advertising services, we expect the conversion advantage this traffic holds over traditional search to grow further.
In the sections below, we break down ChatGPT conversion rates by industry, compare them with each industry’s overall conversion rate, and group the results by market type.
In the table below, we break down the average ChatGPT conversion rate for each of the 32 industries in our dataset, sorted from highest to lowest.
The ChatGPT Conversion Rates by Industry, 2026
| Industry | Average ChatGPT Conversion Rate |
| Hotels & Resorts | 7.3% |
| Legal Services | 5.5% |
| Higher Education & College | 5.2% |
| Entertainment | 4.9% |
| Healthcare | 4.7% |
| HVAC Services | 4.1% |
| Manufacturing | 4.0% |
| Staffing & Recruiting | 3.8% |
| Industrial IoT | 3.8% |
| Pest Control | 3.7% |
| Construction | 3.6% |
| Pharmaceutical | 3.5% |
| Oil & Gas | 3.4% |
| Food & Beverage | 3.3% |
| eCommerce | 3.3% |
| Solar | 3.2% |
| Addiction Treatment | 3.1% |
| Commercial Insurance | 3.1% |
| Real Estate | 2.9% |
| PCB Design & Manufacturing | 2.8% |
| Apparel & Fashion | 2.6% |
| B2B SaaS | 2.6% |
| Medical Device | 2.4% |
| IT & Managed Services | 2.4% |
| Biotech | 2.2% |
| Environmental Services | 2.2% |
| Transportation & Logistics | 2.1% |
| Luxury Goods | 2.0% |
| Software Development | 1.9% |
| Financial Services | 1.8% |
| Heavy Equipment | 1.7% |
| Engineering | 1.6% |
In the table below, we compare each industry’s ChatGPT conversion rate with its overall average conversion rate across all channels, showing the five industries where the gap was largest and the five where it was smallest.
The ChatGPT Conversion Rates: Most vs. Least Improved, 2026
| Industry | Overall Conversion Rate | Average ChatGPT Conversion Rate |
| Hotels & Resorts | 3.4% | 7.3% (+3.9%) |
| Higher Education & College | 2.9% | 5.2% (+2.3%) |
| Entertainment | 3.0% | 4.9% (+1.9%) |
| Manufacturing | 2.2% | 4.0% (+1.8%) |
| Legal Services | 4.0% | 5.5% (+1.5%) |
| Luxury Goods | 1.5% | 2.0% (+0.5%) |
| Biotech | 1.9% | 2.2% (+0.3%) |
| Engineering | 1.4% | 1.6% (+0.2%) |
| Heavy Equipment | 1.6% | 1.7% (+0.1%) |
| Financial Services | 1.7% | 1.8% (+0.1%) |
In the table below, we group the 32 industries in our dataset by their primary buyer (consumer, business, or a mix of both) and report the unweighted average of their ChatGPT conversion rates.
The ChatGPT Conversion Rates by Market Type, 2026
| Market Type | Industries in Dataset | Average ChatGPT Conversion Rate | Highest-Converting Industry | Lowest-Converting Industry |
| B2C | 10 | 4.1% | Hotels & Resorts (7.3%) | Luxury Goods (2.0%) |
| Mixed B2B and B2C | 7 | 3.4% | Legal Services (5.5%) | Financial Services (1.8%) |
| B2B | 15 | 2.7% | Manufacturing (4.0%) | Engineering (1.6%) |
We found that Manufacturing was the only B2B industry to reach 4.0%, consistent with the long, research-heavy partner selection process buyers work through in ChatGPT before contacting a supplier.
We found that B2C industries averaged a 4.1% ChatGPT conversion rate, 1.4 points above the 2.7% B2B average.
Our data showed the widest spread within B2C, where rates ranged from 7.3% to 2.0%, a 5.3-point range compared with 2.4 points across B2B industries.
If you’d like to request a PDF copy of this report or learn more about our GEO services, you can reach out here.
2026-10-29 03:58:20
Last updated October 28, 2026
Between January 2023 and September 2026, our research team analyzed session-level traffic data from 218 client websites spanning 12 industries, covering just over 3.7 billion sessions. The aim of the study was to isolate how much of the change in website traffic across that period is attributable to AI, including AI Overviews inside Google, AI Mode, and referrals arriving directly from assistants such as ChatGPT, Gemini, Perplexity, Claude, and Copilot.
We normalized the panel for the growth of the sites, seasonality, and spend to reflect changes in traffic behavior rather than changes in the client’s budgets. In addition, sites that were added or dropped off the panel in the middle of the year were excluded from this year-over-year analysis. Whenever we talk about AI platform traffic, we mean sessions with referrers that resolve to an AI assistant. On the other hand, “AI Overview” means an answer block that Google inserts above organic results.
Traffic loss was calculated relative to a counterfactual baseline defined by the panel’s pre-2023 growth rate; thus, sites that gained traffic slower than the control group are said to experience a loss, even with an absolute increase in visits. Search intent was categorized according to our own classification framework used in AI content creation research. In the sections below, you will find the breakdown of traffic sources, industries, search intents, and volume of rankings on different platforms.
In the table below, we break down how each traffic source performed across the panel between January 2023 and September 2026. Share of sessions describes the composition of traffic in each period, while change in session volume describes absolute movement after normalizing for panel size.
The AI Impact by Type of Web Traffic, 2026
| Traffic Source | Share of Sessions, Jan 2023 | Share of Sessions, Sep 2026 | Change in Session Volume |
| Organic search | 51.3% | 42.3% | -24.8% |
| Direct traffic | 21.7% | 24.4% | +2.5% |
| Paid search | 11.2% | 13.5% | +9.9% |
| Referral traffic | 8.3% | 7.0% | -23.1% |
| Social traffic | 7.4% | 5.9% | -27.3% |
| AI platforms | 0.1% | 6.9% | +6,193% |
Our researchers took three findings from this dataset:
In the table below, we compare organic and total session change across the twelve industries represented in the panel, alongside the share of each industry’s tracked queries that returned an AI Overview in September 2026.
The AI Impact on Website Traffic by Industry, 2026
| Industry | Change in Organic Sessions | Change in Total Sessions | Queries Returning an AI Overview |
| Media & Publishing | -47.8% | -32.0% | 80.1% |
| Education | -42.3% | -27.3% | 84.6% |
| Healthcare | -38.6% | -24.6% | 87.2% |
| Travel & Hospitality | -35.0% | -20.1% | 71.9% |
| B2B SaaS | -29.8% | -12.9% | 82.0% |
| Professional Services | -28.1% | -11.6% | 69.6% |
| Financial Services | -26.5% | -13.5% | 75.3% |
| Legal Services | -23.0% | -10.0% | 66.9% |
| Manufacturing | -18.6% | -7.3% | 59.2% |
| Ecommerce & Retail | -11.9% | -3.5% | 23.4% |
| Real Estate | -9.5% | -2.9% | 18.8% |
| Local Services | -7.0% | -2.0% | 15.1% |
Three conclusions our team drew from the industry breakdown:
In the table below, we segment the same traffic by the intent behind the query. Every query in the panel was classified into one of four intent categories, then measured for organic session change and for the share of results that ended without a click to any website.
The AI Impact by Search Intent, 2026
| Search Intent | Share of Panel Queries | Change in Organic Sessions | Zero-Click Rate |
| Informational | 46.4% | -44.3% | 92.0% |
| Navigational | 18.6% | -19.7% | 76.7% |
| Commercial investigation | 23.3% | -14.5% | 58.6% |
| Transactional | 11.7% | -5.9% | 37.6% |
Our team took three findings from the intent data:
In the table below, we isolate the cost of an AI Overview at each position on page one. Both columns are drawn from the same set of keywords, measured on impressions where an AI Overview appeared and impressions where it did not.
The AI Impact by Google Ranking Position, 2026
| Ranking Position | CTR Without an AI Overview | CTR With an AI Overview | Relative Change in CTR |
| 1 | 27.4% | 11.6% | -57.7% |
| 2 | 15.1% | 6.9% | -54.3% |
| 3 | 10.3% | 5.0% | -51.5% |
| 4 | 7.6% | 4.0% | -47.4% |
| 5 | 5.8% | 3.4% | -41.4% |
| 6 | 4.4% | 2.7% | -38.6% |
| 7 | 3.5% | 2.2% | -37.1% |
| 8 | 2.9% | 1.9% | -34.5% |
| 9 | 2.4% | 1.6% | -33.3% |
| 10 | 2.1% | 1.4% | -33.3% |
Three findings our researchers drew from the position data:
In the table below, we compare engagement on result pages containing an AI Overview against result pages without one, using the same keyword set and the same date range.
The Traffic Impact of AI Overviews, 2026
| Metric | SERPs Without an AI Overview | SERPs With an AI Overview | Difference |
| Average organic click-through rate | 8.7% | 3.5% | -5.2 pts |
| Zero-click rate | 62.9% | 82.7% | +19.8 pts |
| Organic clicks per 1,000 impressions | 87 | 35 | -59.8% |
| Average session duration after click | 2m 41s | 3m 19s | +23.6% |
| Pages per session | 2.3 | 2.9 | +26.1% |
| Visitor-to-lead rate | 1.7% | 2.6% | +0.9 pts |
Our team took three findings from this comparison:
In the table below, we compare how visitors arriving from each AI platform convert against visitors arriving from Google organic search. Conversion is measured against our standard definition of a marketing qualified lead, consistent with our conversion rate benchmarks by industry and channel.
The AI Traffic vs. Google Traffic Conversion Rates, 2026
| Traffic Source | Visitor-to-Lead Rate | Lead-to-Customer Rate | Avg. Session Duration | Pages per Session |
| ChatGPT | 4.7% | 21.4% | 6m 14s | 3.8 |
| Perplexity | 4.1% | 19.9% | 5m 49s | 3.6 |
| Claude | 3.9% | 22.8% | 6m 44s | 4.1 |
| Gemini | 2.8% | 16.5% | 4m 36s | 2.9 |
| Copilot | 2.4% | 15.2% | 4m 11s | 2.7 |
| Google organic | 1.9% | 13.6% | 3m 20s | 2.4 |
Three conclusions our researchers drew from the conversion data:
In the table below, we track AI platform traffic as a share of all sessions in the panel, quarter by quarter, from the first full quarter after ChatGPT’s public launch through the third quarter of 2026, the last quarter completed inside the study window.
The Share of Website Traffic Coming From AI, 2026
| Quarter | Share of All Sessions |
| Q1 2023 | 0.11% |
| Q2 2023 | 0.15% |
| Q3 2023 | 0.22% |
| Q4 2023 | 0.31% |
| Q1 2024 | 0.44% |
| Q2 2024 | 0.63% |
| Q3 2024 | 0.81% |
| Q4 2024 | 1.29% |
| Q1 2025 | 1.71% |
| Q2 2025 | 2.48% |
| Q3 2025 | 2.96% |
| Q4 2025 | 4.07% |
| Q1 2026 | 5.13% |
| Q2 2026 | 6.24% |
| Q3 2026 | 6.97% |
Our team took three findings from the quarterly series:
In the table below, we break the growth of AI referral sessions down by platform across two full years, alongside each platform’s current share of the AI referrals the panel receives.
The Growth of AI Referral Traffic, 2026
| Platform | Growth, 2024 to 2025 | Growth, 2025 to 2026 | Share of AI Referrals, 2026 |
| ChatGPT | +250% | +76% | 67.9% |
| Gemini | +823% | +203% | 15.0% |
| Perplexity | +307% | +90% | 6.9% |
| Claude | +626% | +205% | 5.5% |
| Copilot | +260% | +199% | 3.1% |
| Other | +382% | +217% | 1.6% |
Three findings our researchers drew from the growth data:
In the table below, we show each platform’s share of the AI referral sessions the panel received in 2026 against its share in 2024. The chart beneath the table traces the same shares quarter by quarter, and the pattern matches what we observed in our agentic AI research.
The AI Platforms That Send the Most Traffic, 2026
| Platform | Share of AI Referrals, 2026 | Share of AI Referrals, 2024 | Change in Share |
| ChatGPT | 67.9% | 84.2% | -16.3 pts |
| Gemini | 15.0% | 4.1% | +10.9 pts |
| Perplexity | 6.9% | 6.8% | +0.1 pts |
| Claude | 5.5% | 1.9% | +3.6 pts |
| Copilot | 3.1% | 2.2% | +0.9 pts |
| Other | 1.6% | 0.8% | +0.8 pts |
Our team took three findings from the platform data:
In the table below, we show how much of each industry’s traffic now arrives from AI platforms. The ordering differs from the organic-loss ordering earlier in this report, and the gap between the two is where we focus most of our B2B conversion research.
The AI Referral Traffic by Industry, 2026
| Industry | Share of Sessions from AI Platforms | Change in Organic Sessions |
| B2B SaaS | 12.1% | -29.8% |
| Professional Services | 9.7% | -28.1% |
| Financial Services | 8.3% | -26.5% |
| Healthcare | 6.9% | -38.6% |
| Education | 6.4% | -42.3% |
| Legal Services | 6.1% | -23.0% |
| Media & Publishing | 5.0% | -47.8% |
| Manufacturing | 4.2% | -18.6% |
| Travel & Hospitality | 3.7% | -35.0% |
| Ecommerce & Retail | 2.4% | -11.9% |
| Real Estate | 2.0% | -9.5% |
| Local Services | 1.5% | -7.0% |
Three conclusions our researchers drew from this breakdown:
In the table below, we track the share of tracked Google searches that ended without a click to any external website, measured annually across the full panel.
The Impact of AI on Zero-Click Searches, 2026
| Year | Share of Searches Ending Without a Click | Year-Over-Year Change |
| 2022 | 54.1% | n/a |
| 2023 | 57.3% | +3.2 pts |
| 2024 | 62.8% | +5.5 pts |
| 2025 | 69.4% | +6.6 pts |
| 2026 | 75.1% | +5.7 pts |
Our team took three findings from the zero-click series:
If you would like a PDF copy of this report, or you would like to learn more about our agency, you can reach out to our team here.
2026-10-29 03:53:32
Last updated October 28, 2026
Between January and October of 2026, our research team studied how AI now resides inside the content-creation process across business, personal communication, and social media. We compiled the dataset from a survey of 1,080 marketing and communications teams, a nationally representative panel of 2,400 U.S. adults, and platform-level sampling of roughly 1.7 million public social posts, then normalized the results so figures are comparable across platforms and company sizes.
For this report, we define AI-generated content as any text, caption, or article where a generative model produced the first draft or a substantial portion of the published wording, whether or not a human later edited it. We separate that from fully human-written content and, where it is useful, break out the degree of human editing applied after the model’s draft.
AI writing tools reached default-utility status faster than most communication technologies before them, and the question our clients ask has shifted from whether to use AI to how much editing published AI content actually needs. This report extends our ongoing research into how generative AI is reshaping content and search. In the sections below, we break the data into three areas: business, personal communication, and social media.
For the business findings, we surveyed 1,080 marketing and communications teams, spanning small businesses, midsize companies, and enterprises, about how they use AI to create content.
In the table below, we break down the share of companies using AI in their content production, segmented by business size.
The Share of Companies Using AI to Create Content by Business Size, 2026
| Business Size | Share Using AI for Content |
| Small Business | 85% |
| Midsize Business | 78% |
| Enterprise | 63% |
The snapshot above is the end point of a fast three-year climb. In the table and chart below, we track the share of each business size using AI to create content from 2023 through 2026.
The Share of Businesses Using AI to Create Content Over Time, 2023 to 2026
| Business Size | 2023 | 2024 | 2025 | 2026 |
| Small Business | 38% | 57% | 73% | 85% |
| Midsize Business | 45% | 58% | 69% | 78% |
| Enterprise | 34% | 45% | 55% | 63% |
Small businesses started slower than midsize companies in 2023 but overtook them by 2024 and pulled ahead through 2026. The crossover reflects how quickly lean teams adopted AI once capable writing tools became widely available. Enterprises trailed for the entire period, gaining ground steadily but held back by procurement cycles and approval structures that slow organization-wide rollout.
In the table below, we compare the average fully loaded cost to produce a standard published article before and after AI adoption, by company size.
The Average Cost Per Article Before and After AI by Business Size, 2022 vs 2026
| Business Size | 2022 (Pre-AI) | 2026 (Post-AI) | Cost Reduction |
| Small Business | $365 | $94 | 74% |
| Midsize Business | $540 | $148 | 73% |
| Enterprise | $720 | $217 | 70% |
These figures represent fully loaded cost, including writer time, editing, and project management, normalized to a 1,000-word article. The reduction is real but not total: larger organizations retain more of their pre-AI cost because compliance review, brand governance, and subject-matter validation do not compress the way drafting does.
Lower cost has not translated into higher quality. In our review, the quality floor rose, since AI drafts rarely carry misspellings or poor grammar, while average quality slipped slightly. We attribute the decline to outsourced thinking, as writers increasingly hand the model the reasoning and structure of a piece rather than only its mechanics.
In the table below, we break down the average time a team spends editing a single piece of AI-generated content before publishing, by company size.
The Average Editing Time Per AI-Generated Piece by Business Size, 2026
| Business Size | Average Editing Time Per Piece |
| Small Business | 4.9 minutes |
| Midsize Business | 11.4 minutes |
| Enterprise | 24.9 minutes |
In the table below, we break down how much editing marketing teams apply to AI drafts, using four tiers we defined by review time.
The Distribution of AI Editing Intensity Across Marketing Teams, 2026
| Editing Tier | Share of Marketing Teams |
| Direct Publish (no human edits) | 9% |
| Light Pass (under 5 minutes) | 36% |
| Standard Review (5 to 20 minutes) | 41% |
| Deep Revision (20 minutes or more) | 14% |
In the table below, we break down what share of a company’s entire website is at least partially AI-generated versus fully human-written, by company size.
The Share of Website Content That Is AI-Generated vs Human-Written by Business Size, 2026
| Business Size | At Least Partially AI-Generated | Fully Human-Written |
| Small Business | 72% | 28% |
| Midsize Business | 64% | 36% |
| Enterprise | 66% | 34% |
Website-level AI share does not move cleanly with company size. Small businesses lead at 72%, since a small site is dominated by a handful of recently produced, AI-touched pages. Midsize companies sit lowest at 64%, old enough to carry a base of legacy human-written pages but not yet producing at the volume that would refresh them. Enterprises tick back up to 66%, where large, continuously updated content operations generate enough new AI-assisted pages to outpace their deep legacy archives.
In the table below, we break down which AI platforms companies rely on most for content creation, by company size. We limited this analysis to the four leading generative AI platforms that accounted for the vast majority of usage in our sample.
The Most Popular AI Platforms for Content Creation by Business Size, 2026
| AI Platform | Small Business | Midsize Business | Enterprise |
| ChatGPT | 53% | 30% | 27% |
| Claude | 24% | 43% | 22% |
| Gemini | 15% | 19% | 43% |
| Grok | 8% | 8% | 8% |
The shares above are an October 2026 snapshot of a fast-moving market. In the table and chart below, we track each platform’s blended share of content-creation usage month by month across 2026.
The AI Platform Share of Content Creation Over Time, January to October 2026
| AI Platform | Jan | Feb | Mar | Apr | May | Jun | Jul | Aug | Sep | Oct |
| ChatGPT | 41% | 40% | 37% | 35% | 33% | 32% | 32% | 39% | 37% | 36% |
| Claude | 24% | 26% | 29% | 32% | 34% | 36% | 35% | 28% | 30% | 31% |
| Gemini | 30% | 29% | 28% | 27% | 27% | 26% | 26% | 26% | 26% | 25% |
| Grok | 5% | 5% | 6% | 6% | 6% | 6% | 7% | 7% | 7% | 8% |
The monthly view shows a genuine race for the top. Claude climbed steadily through the first half of 2026 and passed ChatGPT in the late spring, peaking in June and July. In August its share fell back sharply as ChatGPT regained the lead, a swing large enough that a single-month reading can misrepresent the trend. September and October showed a partial settling, with ChatGPT easing to 36% and Claude recovering to 31%, though ChatGPT held the lead through October. Gemini drifted down gently across the year, from 30% to 25%, while Grok held the bottom position throughout but gained a little ground, from 5% to 8%.
In the table below, we break down the estimated time saved per published piece when AI is involved versus a fully human workflow, by company size.
The Estimated Time Savings Per Published Piece by Business Size, 2026
| Business Size | Time Saved (%) | Hours Saved Per Piece |
| Small Business | 62% | 3.4 hours |
| Midsize Business | 66% | 4.2 hours |
| Enterprise | 59% | 3.8 hours |
Savings peak at midsize companies on both measures: they save 66% of production time and 4.2 hours per piece, more than either smaller or larger firms. Small businesses save a large fraction of an already light process, so the absolute hours come out lower. Enterprises run the longest pieces but return part of their potential savings to layered review and approvals, the same overhead reflected in their editing times above, which holds both their percentage and their hours below midsize.
In the table below, we break down what share of content ranking in the top Google results is AI-generated, separated by how much human editing the content received.
The AI-Generation Level of Content Ranking in Top Google Results, 2026
| Level of AI Generation | Share of Top-Ranking Content |
| Entirely AI, minimal to no human editing | 0.4% |
| Partially AI, 5+ minutes of human editing per piece | 58.3% |
| Fully human-generated | 41.3% |
In the table below, we break down how businesspeople react when they detect that content sent to them was AI-generated.
The Business Reaction to Detected AI-Generated Content, 2026
| Reaction | Share of Businesses |
| Slightly positive | 4% |
| Slightly negative | 49% |
| Moderately negative | 34% |
| Very negative (refuse communication or transaction) | 13% |
For the personal findings, we surveyed a nationally representative panel of 2,400 U.S. adults about how they use AI to create content in their own lives.
Here, using AI to create content for others means drafting or generating the wording of a message a person then sends, most often an email, a direct message on social media, or a text on a service such as WhatsApp or Telegram. In the table below, we break down the share of U.S. adults who do this, segmented by age group.
The Share of U.S. Adults Using AI to Create Content for Communication by Age, 2026
| Age Group | Share Using AI to Communicate |
| 18 to 29 | 69% |
| 30 to 44 | 58% |
| 45 to 60 | 39% |
| Over 60 | 22% |
| All U.S. adults | 46% |
In the table below, we break down the most common personal use cases among adults who use AI to create content.
The Top Use Cases for Personal AI Content Creation, 2026
| Use Case | Share of AI-Using Adults |
| Resumes and cover letters | 75% |
| Professional emails and messages | 59% |
| Social media posts and captions | 45% |
| Customer service messages and complaints | 31% |
| Personal notes, cards, and letters | 28% |
| Online dating profiles and messages | 22% |
| Reviews and feedback | 19% |
In the table below, we break down how individuals react when they detect that a personal message they received was AI-generated.
The Personal Reaction to Detected AI-Generated Content, 2026
| Reaction | Share of Individuals |
| Slightly positive | 3% |
| Slightly negative | 38% |
| Moderately negative | 40% |
| Very negative (refuse communication or transaction) | 19% |
Individuals reacted more harshly to AI in personal settings than businesspeople did in professional ones. The moderately negative and very negative tiers together reached 59% among individuals versus 47% among businesspeople, and only 3% of individuals viewed detected AI use positively. Both patterns are consistent with an expectation that personal communication carries a signal of genuine effort that AI is seen to undercut.
For the social media findings, we sampled and classified roughly 1.7 million public posts across the platforms studied to estimate how much of their content is AI-generated.
In the table below, we break down the share of publicly posted content on each major platform that our sampling flagged as at least partially AI-generated in 2026.
The AI-Generated Share of Social Media Content by Platform, 2026
| Platform | AI-Generated Share of Content |
| 56% | |
| X (formerly Twitter) | 43% |
| 39% | |
| 35% | |
| 30% | |
| YouTube | 27% |
| TikTok | 24% |
| 18% |
The snapshot above reflects a steep climb. In the table and chart below, we track the AI-generated share of content on five platforms from 2021 through 2026.
The AI-Generated Share of Social Media Content Over Time, 2021 to 2026
| Platform | 2021 | 2022 | 2023 | 2024 | 2025 | 2026 |
| 7% | 13% | 24% | 33% | 46% | 56% | |
| X (formerly Twitter) | 5% | 9% | 18% | 26% | 35% | 43% |
| 6% | 10% | 16% | 25% | 32% | 39% | |
| 4% | 7% | 13% | 18% | 24% | 30% | |
| TikTok | 2% | 6% | 11% | 15% | 19% | 24% |
The trajectories are not uniform. LinkedIn and X pulled away from the pack after 2022, coinciding with the mainstream release of general-purpose writing assistants, while Instagram and TikTok grew more slowly off a smaller base. We modeled each platform’s year-over-year change independently rather than fitting a single curve, which is why the lines cross: X overtook Facebook in 2023 as its text-post volume shifted toward AI drafting, and the two moved in near-parallel afterward.
If you’d like to request a PDF copy of this report or learn more about how our agency approaches content and search strategy, you can reach out here.