← All issues

The GTM Tech Index Brief

September 27 to October 18, 2026 · Published October 18, 2026

Three weeks in one line

About thirty GTM vendors plugged into AI assistants in three weeks, and the better ones shipped a ceiling with the plug, so an outside agent can read the pipeline without rewriting it. Outreach, Clay and Octave started keeping the company story in a record their agents read before writing, so the messaging doc finally has a reader who opens it every time. And AiSDR priced an AI SDR at $1.50 per person contacted, which gives finance a unit it recognizes.

This issue covers the three weeks from September 27 to October 18, with 109 entries across 109 vendors, one each. Eighty two were verified at the vendor’s own source and twenty seven through other reporting. Integration work accounted for thirty of them, and forty one were rated high impact.

Most of it circled one question from several directions: what software should be allowed to do while nobody is looking.

Everything got an MCP server

More than a quarter of the period’s entries put a GTM product within reach of an outside AI assistant, through an MCP server, a connector or a listing in an assistant’s app directory.

An MCP server used to be a launch. Over these three weeks it became something closer to a footer link, expected everywhere and noticed mainly when it is missing.

Where the connector lives started to matter as much as what it does. Hunter launched an app in the ChatGPT apps directory with more than 30 tools, included on every plan down to Free. Leadfeeder put a plugin in the Claude and ChatGPT marketplaces that bundles its MCP server with six prewritten skills, among them a daily brief on the companies visiting your website.

Airspeed listed its MCP server in the Claude and ChatGPT directories, and Consensus listed in Claude’s and Gong’s. Evergrowth made its Eva agent available in Claude, ChatGPT and Microsoft 365 Copilot, and Reachdesk documented its server inside Microsoft Copilot Studio so gifting can be requested from Teams. Both are recorded Partially Verified.

Sendoso went to the data team instead, listing on Databricks Marketplace so a lead score can trigger a gift without a CSV ever changing hands.

Slack turned into an assistant surface of its own. Vidyard and GetAccept both work through Slackbot now, and GetAccept’s version can send signing reminders to buyers from inside a thread. Tuco AI mirrors each lead’s text conversation into a Slack thread, where a reply goes back to the lead as a text.

Most of these connectors draft rather than send. GetResponse, Expandi and Outboundy leave the assistant’s work as drafts for a person to review, with Expandi and Outboundy recorded Partially Verified.

A few go further. A connected assistant can validate and launch a Skylead campaign, and the API key carries that authority, recorded Partially Verified. Dripify, also Partially Verified, and DM Faster opened LinkedIn campaigns and social lead extraction to agents, along with the platform terms questions that come attached.

Our read

The MCP server has gone the way of the Salesforce integration a decade ago, from differentiator to entry ticket. The assistant’s directory is becoming the new marketplace page, and a listing there is distribution. What still separates products is the line between draft and send, and whether a buyer can see where it is drawn before connecting anything.

The permission slip arrived

The more interesting work sat on the far side of the connector, in the controls that decide what an outside agent may do once it is in.

Fullcast added a per user ceiling with four levels, from read only tools to full access. An agent asked to go past it reports the action as unavailable rather than hunting for a workaround, which is a better instinct than most of us show at a locked door. Recorded Partially Verified.

SPOTIO split its AI permissions by role. Chat is one switch, AI Actions, which let the assistant change records, is another, and connecting outside AI clients is a third. A rep can have an assistant that answers questions without one that edits the account.

Admins also got a view of who is connected. Revenue Grid added a page listing every application that reaches its data over MCP, with publisher verification, recent usage and how many users authorized each one. Planhat lets admins wall off fields, records and named customers from its Pi agent, turning a common enterprise contract condition into a setting.

OutreachBox shows an agent only the tools its API key is scoped for, so a read only key never sees a send tool. Keys created before scopes existed count as unrestricted, which is the detail worth checking first. Gainsight opened community moderation and publishing to assistants for staff roles, and kept a confirmation on the one action that cannot be undone.

Not every guardrail sits on the vendor’s side. Notta’s remote MCP server can permanently delete records, and its guide asks client developers to get the user’s approval for each one, recorded Partially Verified. That is a rule written for the client rather than a check the server enforces.

Sybill went the other way. Its scheduled prompts can now act through MCP connectors, and the release note says the selected tools run with no approval prompts, including sending the follow up email as the rep.

Our read

The connector is table stakes now, and the permission model is the product. The vendors worth watching answer three questions in the admin console rather than the docs: what an outside agent can see, what it can change, and how anyone finds out afterward. Expect those three to join the security review, right after the request for the SOC 2 report.

Autonomy, one notch at a time

The vendors’ own agents got more capable, and several shipped the brakes with the same care as the engine.

LeanData introduced Run, an agent for revenue operations that proposes fixes for approval, cites the rule behind each answer, and takes on a type of decision by itself only after repeated approvals. In that design, autonomy is earned one decision type at a time rather than granted on install day. Recorded Partially Verified.

MarketBetter launched Loops, a nurture agent built to send less. It writes only when a dated, sourced update fits the person, and a second model from a different model family must approve each draft or the email is blocked. An AI whose main skill is deciding not to email someone may be the boldest product in this issue.

Salestools shows an SDR agent’s rules before launch and locks the ones protecting customers, unsubscribes, Do Not Contact lists and conversations handed to a person. OutX documented an agent that retunes LinkedIn listening watchlists against what they actually collect. It is opt in, never acts within a week of a person’s edit, and can be reverted. Recorded Partially Verified.

Confirmation steps showed up across the stack. Pipeline CRM opened Copilot in beta with a Strict mode an admin can enforce for the whole account, and Copilot cannot send email or touch the dialer. ActiveCampaign’s assistant previews every reply and every flow it activates, and flows it drafts start paused.

SuperSend lets an agent take an account from setup to launch while payment and the final launch check stay with a person. Clay’s new global agent and Sendr’s Campaign Copilot both put a plan in front of the user before any outreach starts.

The other end of the dial moved too. lemonreach can run its website visitor monitors fully automated, recorded Partially Verified, and SalesBlink’s new reply agent can answer leads on autopilot. Topo’s agent now runs tasks on a schedule, also Partially Verified.

Our read

Autonomy is turning into a dial with named positions rather than a yes or no. That is good product design, and it changes the comparison. The useful question about two agents is no longer what each can do. It is what each does by default on day one, and what it takes to move it up a notch.

Positioning became an input

Four launches turned a company’s own story into something its agents read before they write.

Outreach added Org Context, admin curated blocks on the company, its competitors and its products that every Outreach agent consults before generating anything. Admins can fill the blocks from web sources, accept or reject each suggestion with its sources shown, and archive what agents should stop using.

Clay’s global agent reasons from Knowledge Hub, a stored record of the customer’s ideal customer profile, personas and positioning. Octave shipped Connections, which checks linked Notion pages, Drive files and live webpages every week and flags meaningful differences. An optional two way sync lets Octave edit those documents when its own library changes.

Apollo.io launched an Intelligence Layer that merges its data with a customer’s CRM and engagement history, then decides which accounts to work and what to do next. Apollo says it acts with minimal human involvement, which makes the quality of what it reads first rather important.

Product marketing has spent years asking people to read the messaging doc. The first reader to open it every single time turns out to be an agent.

Our read

The competitors block in an admin console is now what every agent repeats to a prospect. That moves positioning from a slide to a system of record, with what systems of record need: an owner, a review cycle, and a way to retire a claim before an agent says it to a buyer. Two way sync adds a fair question for any team that adopts it, which is whose document wins when the marketer and the machine disagree.

Everyone wants to sell you the contact

Gong made Gong Enrich generally available, bringing outside contact and account data into its platform with launch partners including Apollo, ZoomInfo, Lusha, LeadIQ, RocketReach and ZeroBounce. Teams can spend credits they already hold or buy more through Gong, and RocketReach went live as a provider in the same weeks.

A conversation intelligence platform now sells the data that many of its customers buy separately from enrichment tools. FoxReach moved the opposite way, adding a Lead Finder so a sending tool supplies the contacts too. Its release does not describe the licensing behind the profile data.

Provenance got more granular at the same time. Lusha now stamps every email and phone number in an enrichment response with its source, Lusha or the outside provider in its waterfall. Coresignal removed three professional email fields from its employee API and pointed customers to a separate contact enrichment product.

Seamless.AI put intent filters into search and states plainly that intent is measured at the company, so a contact shown with intent reflects activity at the employer rather than by the person. That one sentence will spare a number of awkward first calls.

Signals kept multiplying. Signado and Surfe both turned job postings into a lead source, Bombora added 3,768 intent topics for a total of 25,378, and RB2B launched an API that pulls up to 90 days of identified website visitor profiles. Demandbase now shares its data into a customer’s own Snowflake account without copying it.

Our read

A contact can now be bought inside a call recorder, filled through a waterfall of providers and sent from an assistant, and each hand it passes through has its own terms. Lusha’s per field source is the right shape for that world, because it is the one piece of provenance that travels with the record. The count at the end of this issue shows how many data vendors answer the question at all.

The recorder started asking first

Calls and meetings got the same treatment as agents: more capability, and a sharper question about who agreed to it.

MaxIQ now emails every participant before its recorder joins and skips the recording if anyone declines. That is consent rather than notice, and it fits all party consent states far better than a banner does.

Two vendors took the bot out of the meeting. Salesloft can import Microsoft Teams’ own cloud recordings, so participants see the Teams recording notice rather than an outside bot joining. Reevo’s new mobile app records in person meetings and calls without one, recorded Partially Verified, which moves the duty to tell people onto the seller holding the phone.

Granola’s Snapshots capture slides and documents other participants share, switched off by default on Enterprise plans and with consent reminders when capture starts. Storylane lets users clone their own voice for demo narration, which leaves a buyer listening to a synthetic voice unless someone says so.

The AI caller arrived in earnest. Kixie’s Follow Up AI Agent calls a contact after a chosen call outcome, says it is an AI assistant, offers an opt out, stays inside a calling window, and cannot go live until a manager records a consent attestation. Recorded Partially Verified.

Artisan’s agent, Artie, now has its own phone number for calls in both directions, still behind access codes. The release does not say whether it tells people they are speaking with an AI.

Dialing got faster and better aimed. Smartlead added auto dialing, without describing how calling hours or do not call lists are handled. TitanX launched a single line dialer where a person places every call, with monitoring that pulls and repairs numbers flagged as spam. Aircall brought branded caller ID to the UK, Germany, France and Spain, and folk started placing and transcribing calls from the user’s own number.

Our read

Taking the bot out of the meeting is good for the meeting and harder on consent, since the bot was often the most visible sign that anyone was recording. The index has grown from 314 vendors to 408 since the September 5 issue counted four that document their automated outreach discloses itself. The count is still four. Kixie’s agent shows the full version can be built: disclosure, opt out, calling windows and an attestation, set up inside the product rather than left to the terms of service.

Pricing moved to the result

AiSDR replaced its monthly plans, which ran from $250 to $2,500, with pay as you go pricing at $1.50 per person contacted, charged daily with no contract. Above 1,000 contacts a month, sales led Team plans take over at up to 40 percent less per contact.

An AI SDR priced per person contacted gives finance a unit it can set beside a human SDR’s cost per touch. Somewhere, a spreadsheet just gained a very interesting column.

Others priced the outcome too. Generect now charges for an email lookup only when it confirms an address, recorded Partially Verified. Autobound retires its older Insights API on October 30, and the Signal API that replaces it charges 2 credits per returned signal rather than per request.

Credits also spread further into the base plan. HubSpot relaunched Starter with its Prospecting Agent, Notetaker and meeting prep included, with agent runs drawn from HubSpot Credits. Snov.io folded a monthly quota of enrichment tokens into every plan at no change in price.

Zoho CRM lets admins switch on OpenAI or Anthropic models without their own API keys, metered through Zoho’s credits, which simplifies procurement and routes CRM data to those providers through Zoho. Manychat passed through Meta’s longer WhatsApp windows for click to WhatsApp ads, giving businesses seven days instead of 24 hours to answer, followed by a free seven day messaging window.

Our read

The seat is giving way to the result: a person contacted, an address confirmed, a signal returned. That is easier to budget per unit and harder to forecast in total, because the bill now scales with how hard the agent works. For anyone pricing an agent, the first question in every deal is becoming what one unit of work costs, and who decides how many get spent.

The quiet fixes

Opt outs started traveling between systems. Aloware now writes opt out status into HubSpot, so a number that said stop in one tool is not texted from another. Landbase matches Do Not Contact entries against company website domains, so a listed company is also recognized in dialer and LinkedIn work.

Sending discipline got more precise. Salesforge now pauses only the LinkedIn action that hit a limit, on only that account, for six hours. InstaOutreach starts new Instagram accounts at 2 DMs a day and adds 2 a day up to a chosen cap. Enginy lets one campaign send from a pool of mailboxes or LinkedIn accounts, recorded Partially Verified, which raises throughput and is also the pattern providers watch for.

Instantly added a Deliverability Agent that checks sending setup and can help fix what it finds, which puts an agent next to the settings that decide inbox placement. Primeforge opened bounce and blacklist health to its API, so agencies can watch a mailbox fleet from their own dashboards.

Two releases took on the paperwork behind outbound calling. Saleshandy added a tab for complaints under India’s rules on unsolicited commercial communication, with proof of consent uploaded per complaint. LeadDesk lets contact centers verify a business address once and reuse it across outbound numbers.

One default moved the other way. QuickProsp changed the opt out on new cold campaigns to a reply asking to stop, and leaves out the bulk mail unsubscribe headers that mailbox providers use to file mail as promotions. Existing campaigns keep their settings.

A reply to stop request can meet the CAN-SPAM opt out rule. Gmail and Yahoo, however, require one click unsubscribe headers from bulk senders, so a higher volume team should check the new default before relying on it.

Our read

Most of this section is restraint shipped as a feature, and none of it will headline a launch event. The QuickProsp change is the one to think through. A default that trades the unsubscribe header for a plainer looking email is a reasonable bet for a small sender and a risky one for a large sender, and only the buyer knows which they are. Defaults are product decisions with compliance consequences.

Evidence with the method showing

Spear, the new name for Samplead, ranked nine trigger types across 177,629 LinkedIn connection requests from more than 90 teams over eleven months, with internal and test accounts excluded. Podcast and interview triggers did best, at 46 percent acceptance and an 18 percent reply rate.

The first issue of this Brief noted a vendor turning podcast appearances into a buying signal. The data suggests it was onto something.

The decay figure may be the more useful one. Acceptance on LinkedIn activity triggers fell from 44 percent within a week to 31 percent after 60 days. Signals are perishable, and this one loses close to a third of its pull in two months.

Rox published a model that scores cold emails, trained on drafts before and after SDR edits. On 717 held out edits it preferred the edited version 80 percent of the time, against below 50 percent for GPT and Claude used as judges. Rox uses it to pick among drafts, and showing how an agent chooses which email to send is rare in this category. Recorded Partially Verified.

Three customer results rounded it out, all on the vendors’ own figures. UserGems reported five times the lead creation at Quickbase against its 2025 average quarter, with reps approving every message. Dashly’s chat agent booked 78 percent of FitBase’s meetings in September.

Quantified AI described a Novartis oncology launch where every rep certified through AI roleplay inside a window shortened by about three months, recorded Partially Verified.

Our read

A study with a sample, a period and stated exclusions is still the vendor’s own study. It is also something a buyer can argue with, which is the point. The category’s evidence problem was never a shortage of numbers. It was numbers with nothing attached, and these three weeks produced a handful with the method showing.

Market notes

Language support widened. Dialpad extended its full AI feature set to eight more languages in two phases running into late November, and Fathom 3.0 writes summaries in English plus 38 more, on Mac first, recorded Partially Verified.

Several platforms got easier to run inside a larger company. 6sense released a self service sandbox for testing audience and data workflows before they touch production. Close gathered single sign on, two factor enforcement and support access controls onto one security page. Cirrus Insight joined the Cloud Security Alliance STAR registry at Level One, a self assessment rather than an audit.

On data handling, PeopleLinx published a revised data processing agreement with a written commitment not to train general purpose models on customer personal data. MeetGeek can file meeting records to SharePoint under the customer’s own retention rules, and Odie added Microsoft Fabric as a data source under access the customer scopes. All three are recorded Partially Verified.

Leaving got slightly easier in two places. GTM Buddy can export a course as a SCORM file for another learning system, and Attio announced a read only, Postgres compatible endpoint that BI tools can query directly. Both are recorded Partially Verified.

More agents arrived inside products people already use. ChurnZero introduced Kit, which answers questions about accounts and drafts outreach that waits for review, recorded Partially Verified. Seismic added a Deal Agent, and a Library Agent that recommends content changes for approval before it updates anything.

Dock now builds deal rooms, portals and order forms from a prompt, quotes included. Act-On’s Liveboard Agent builds campaign reporting from a typed request, and Outdoo AI put roleplay and real calls in one interface. Avoma turned recurring questions into scheduled reports, recorded Partially Verified.

Fireflies launched voice dictation across the desktop, free on every plan, and Instantly rebuilt its app for phones and tablets. Founders who answer leads from the school pickup line have been heard.

Connections kept deepening. Boomerang AI shipped a public API for its warm introduction data, and Salesmotion writes 35 account signal properties into HubSpot, recorded Partially Verified. Woodpecker can now answer LinkedIn replies through its API, and Membrain, OnePageCRM and Read AI all widened what a connected assistant can do.

At events, Captello shows an account brief the moment a lead is scanned, and Togile added support tickets to its CRM. Since September 30, Marketo Engage rejects activity calls against static lists of 10,000 leads or more, which someone’s nightly sync job is about to discover.

What three weeks say about the space

A hundred and nine entries, one per vendor, and the through line is plain. The GTM stack spent three weeks making itself reachable by machines, then deciding what those machines may do once they arrive.

The connector stopped being news. The permission model, the approval step and the stored company story became the parts worth comparing. Several vendors staged autonomy rather than switching it on, and at least one shipped an agent whose main skill is restraint.

Underneath, the record of where things come from got harder to keep. A contact can be bought inside a call recorder and sent from an assistant. A meeting can be recorded with no bot in the room. The vendors stamping a source on every field and asking every participant before recording are building for that world.

The count below shows how many data vendors say where their contacts come from at all.

Index Answer

Which B2B data vendors say where their contacts come from?

Fewer than half. Of the 143 vendors in the GTM Tech Index that sell or resell B2B contact data, 54 document their sourcing and licensing at a standard a buyer can check, 38 percent.

Six do it in full, naming the inputs, the terms the data arrives under and who stands behind the answer: Apollo.io, Bombora, GTM Workspace by ZoomInfo, Instantly, Leadfeeder and Saleshandy.

Another 48 document their sourcing with at least one material question left open, such as the license behind the data or who stands behind it. Of the remaining 89, 74 describe their data in general terms and 15 do not address the question at all.

Where sourcing is disclosed, it tends to come from four places. There is public web collection, data contributed by users of a community or browser extension, licensed third party files, and contribution networks where customers trade their own records for access to everyone else’s. Each carries different obligations for the buyer, which is why the disclosure matters more than the method.

The past three weeks show why the question is getting harder. A contact bought through a call recorder’s enrichment waterfall comes from one of several providers, each with its own terms. A field stamped with its source, as Lusha’s now are, can be checked. One that arrives unstamped hands the checking to the buyer, usually after the sequence has already run.

The vendors that publish a sourcing statement, and the artifact each grade rests on, are on the data provenance page.

The GTM Tech Index Brief is published by GTM Tech Index, an independent reference for evaluating the software revenue teams use to find, win, and keep customers. No vendor pays for inclusion, placement, or rating. Compare any indexed vendors by capability at Compare and read the evaluation standards at Methodology.

Contact us

Found a vendor we missed? Have feedback on the index? We’d love to hear from you.