The "98% SMS open rate" is repeated on almost every text marketing page, but we could not trace it to a primary study, and it cannot be measured. Plain SMS has no read tracking, so no provider knows how many texts are actually opened. The figure is best read as a loose description of delivery, and it should not be the basis for a business case for an AI texting agent.
This matters if you are deciding whether to put an AI agent on SMS. A made-up benchmark sets expectations no deployment can be checked against. The rest of this page covers what we could find about the number and what to measure instead.
Where does the 98% figure come from?
Nobody agrees. Marketing blogs credit at least four different sources: Gartner, a 2010 mobile advertising white paper, an FCC advisory committee report and the Mobile Marketing Association. Some of these pages contradict each other, and links that once pointed to the supposed original no longer resolve to it. We could not find a primary document that confirms any of the attributions.
The scepticism is not new. Burst SMS, an SMS marketing company, asked how accurate the 98% open rate is back in 2015. More recent roundups, such as Carly's SMS statistics list, note that the number is untraceable even as it keeps being cited.
Why can't anyone measure SMS opens?
Email open rates rely on a tracking pixel: a tiny image that loads when the email is opened and tells the sender it was viewed. SMS has no equivalent. A plain text message is just text, with nothing that reports back when the recipient looks at it.
What providers do report is delivery status. When you send a message through a platform such as Twilio, you get status updates telling you whether it was delivered to the handset, failed or went undelivered. Twilio even charges $0.001 for each message that ends in a failed status, as of November 2026. Delivery is real and measurable. Opening is not.
| Channel | Delivery status | Read tracking |
|---|---|---|
| Plain SMS | Yes, from your provider | No |
| MMS | Yes, from your provider | No |
| RCS | Yes | Read receipts possible if the recipient allows them |
| iMessage | Yes | Read receipts possible if the recipient allows them |
| Partial | Pixel-based, and increasingly blocked |
So "98% open rate" most plausibly means something like "almost all texts get delivered and most people glance at their texts". That may well be true for many senders, but it is a belief about behaviour, not a measured rate.
Do customers actually want business texts?
The surveys are more useful than the 98% figure, but they disagree, and most come from companies that sell texting:
- Business-message categories where texting was preferred (EZ Texting, n=959 US, Jan 2026)
- 90%
- Respondents opted in to some business texts (SimpleTexting, n=1,000, Jan 2026)
- 86%
- Receive business SMS (MEF 2026 global survey)
- 37%
- Prefer business SMS (MEF 2026 global survey)
- 21%
The EZ Texting report and SimpleTexting's survey are US-focused and come from texting vendors. The MEF survey is global and gives a cooler picture: far fewer people prefer business SMS than receive it. Both can be true. US consumers who have opted in to texts from a business they use may like them a lot, while people in markets where WhatsApp dominates may see SMS mainly as a channel for codes and alerts.
For a business choosing a channel for an AI agent, the takeaway is that preference depends on market and context. Our channel guide and WhatsApp business agent guide cover the alternatives.
What should you measure for an AI texting agent?
Skip opens. Measure what your own logs can show:
| Metric | What it tells you | Where it comes from |
|---|---|---|
| Delivery rate | Whether your registration, numbers and content are getting through carriers | Provider status callbacks |
| Reply rate | Whether customers engage with the agent | Inbound message logs |
| Opt-out rate | Whether messages are unwanted or too frequent | STOP and similar replies |
| Time to first reply | Whether the agent responds fast enough | Timestamps on inbound and outbound |
| Resolution rate | Whether the agent actually solves the problem | Conversation outcome tags |
| Handoff rate | How often a person has to step in | Escalation events |
- Log every status updateStore delivery, failure and undelivered events against each message.
- Tag outcomesMark each conversation as resolved, handed off or abandoned.
- Track opt-outs separatelyA rising opt-out rate is the clearest sign something is wrong.
- Compare against your baselineUse your own pre-AI numbers, not industry averages.
A falling delivery rate usually points to a registration or content problem. Our 10DLC registration guide covers the most common causes. A rising opt-out rate means the agent is texting too often or saying the wrong thing, and every opt-out must be honoured, as set out in our opt-in and opt-out rules.
Resolution rate also drives cost if you use a platform that charges per outcome. Intercom's Fin, for example, charges $0.99 per outcome, as of November 2026. Our SMS AI agent cost guide has the other per-resolution prices.
Should you avoid SMS statistics altogether?
No, but be selective. Some figures are well sourced and useful:
- Prices published by providers, such as Twilio's $0.0083 per US SMS segment.
- Regulatory figures, such as the FTC's report of $15.9 billion in fraud losses in 2025, which is relevant to how wary customers are of texts.
- Named surveys with a sample size and date, read with the publisher's interests in mind.
Figures with no source, no date and no method, like the 98% open rate, are best left out of business cases. If you need to argue that texting gets attention, say that most people keep their phones close and that delivery rates for registered traffic are high, and then prove it with your own delivery and reply data. For UK businesses weighing texting against other channels, personalagent.uk's small business guide covers the UK rules.
Bottom line
The 98% SMS open rate is a commonly cited, unverified figure. We found no primary source, and plain SMS cannot report opens, so the number describes delivery and habit at best. Consumer surveys give a mixed picture: strong preference in some US vendor surveys, much weaker in MEF's global data. For an AI texting agent, measure delivery, replies, opt-outs, resolution and handoffs on your own traffic. Those numbers are real, and they tell you whether the agent is working. For use cases where texting tends to work well, see our guides to AI customer support over SMS and AI text reminders.


