Everyone uses it, nobody owns the record it reads

Two datasets landed this week that look unrelated and are not. On September 2, Salesloft published its 2026 Revenue Benchmark Report, US Edition, in which all 500 US sales and revenue decision makers surveyed reported using AI somewhere in the revenue process, and 20.6% called their AI strategy production-ready with measurable outcomes. On September 3, Kyle Poyar and Profound published a study of 7,600 AI responses about pricing across the Cloud 100, and found that a company's own pricing page was cited first in 12% of responses. Both are vendor-run studies and we treat the figures as claims. The shape is the same either way.

In one case the machine is acting on a record the company controls but has not maintained. In the other the machine is describing the company from records it does not control at all. Salesloft's respondents said CRM updates are the top administrative bottleneck at 37.6%, and that 55.6% of loss information rests mostly on subjective seller reporting. Poyar and Profound found that the three domains AI leans on hardest for pricing are Vendr at 18.7% of runs, Reddit at 18.6%, and G2 at 15.9%. Different systems, same failure: the evidence a model reasons over was assembled by somebody else.

The third item of the week makes the point physically. Profound's research team found that on July 10, one day after OpenAI shipped GPT 5.6 models, ChatGPT Shopping flipped from pulling product recommendations mainly from web search to pulling them mainly from its own integrated product feed, moving from 8.26% to 61.54% in a day. Nothing a merchant wrote changed. The retrieval path changed. Visibility that was earned through content became visibility that requires an integration.

The useful question this week is not what your AI can do. It is a simpler one: for each decision you are letting a model make or influence, name the record it reads, and name who is responsible for that record being right. In most organizations, at least one of those two answers is missing.

Every item below carries a source link. Where a figure comes from a vendor describing its own product, its own customers, or its own survey panel, we label it a claim and attribute it.

The big picture

Three shifts to take into your next pipeline call

AI adoption is now a useless signal, so stop measuring yourself against it. Salesloft reports 100% adoption and 20.6% production readiness across 500 US revenue leaders (GlobeNewswire, September 2, claim). Every company will answer yes to "are you using AI." The question that separates them is whether they can name a workflow where an AI recommendation is logged, coached against, and tied to a number. Four out of five cannot, on their own vendor's data.
Answer engine optimization is now a Forrester category, and Forrester has already set the bar for the next round. Forrester published its Answer Engine Optimization Technologies Landscape, Q3 2026 on August 28, creating a named category, and wrote that the next frontier is correlating visibility and sentiment to outcomes like sales (Profound, September 4). Category creation is what makes AEO a budget line. It is also what makes "we improved your citation share" stop being enough by the next planning cycle.
Retrieval changed under everyone's feet, and content did not cause it. Profound recorded feed-integrated retrieval in ChatGPT Shopping rising from 8.26% to 61.54% in a single day on July 10, generalized across roughly 1.75 million shopping prompts, and says brands without an integrated feed are now out of consideration for about 65% of product recommendations (Profound, September 3, claim). If you sell product, discoverability inside an assistant is now partly an engineering ticket, not a content brief.

RevOps

Universal adoption, unprovable results

Benchmark

Salesloft says AI use hit 100% among the revenue leaders it surveyed, and 20.6% call it production-ready

Salesloft 2026 Revenue Benchmark Report, US Edition, announced September 2 via GlobeNewswire. Operator read-through via MarketScale, September 6. n = 500 US sales and revenue decision makers.

Vendor claimAll 500 respondents reported using AI somewhere in the revenue process. Only 20.6% said their AI strategy is production-ready with measurable outcomes, and 28.2% said they are still experimenting. On autonomy, 38.4% preferred a guided model where AI can recommend or act while humans keep oversight.

The operational numbers underneath are the interesting part, and they are not about AI at all. Updating CRM records was the top administrative bottleneck for 37.6% of respondents, and 31.4% named manual CRM administration as the biggest barrier to pipeline generation. While 84% said loss reasons are captured often or always, 55.6% said those entries rely mostly on subjective seller reporting. Only around 32% said they can immediately pinpoint why a deal stalled, and 27% said they can see win-loss rates but cannot explain what happened between stages. Meanwhile the top 10% of sellers generated 47.4% of closed-won revenue, average quota attainment sat near 62%, and 68.4% of leaders reported higher pipeline quotas.

On stack strategy, 26% said they are actively consolidating revenue technology, 32.6% are evaluating where consolidation makes sense, and 26.4% still prefer specialized point solutions.

Why it matters: This is the cleanest single-source argument we have seen this year for fixing data and process before buying more AI. Most teams are already paying for the models. What they cannot do is explain a stalled deal, and the reason is that stage-change evidence lives in a free-text field a seller filled in from memory. That is fixable, unglamorous work, and it is the prerequisite for everything bought last quarter.

Hiring

The GTM engineer title keeps expanding, and the tracking of it is getting noisier

GTME Pulse jobs archive and job market analysis, figures read September 7, 2026.

GTME Pulse reports 177 active qualified listings as of September 7, 2026, against a dated archive of 65 qualified listings with 36 usable pay records from a May 4 export, median posted range midpoint of $150K. Its market analysis puts posting growth at 5,205% from 2024 to 2025, from 63 to 3,342 postings. Worth flagging honestly: the same page also states 19 active listings as of August 31, 2026, so read the active count as directional rather than precise.

The taxonomy on that page is more useful than the counts. It draws a line between roles that own revenue systems, automation, data, integrations and workflows, and roles that carry the title but no technical remit. Posted ranges in the archive spread from $50K to $80K at the bottom to $250K to $330K for director-level GTM engineering at Airwallex and Altruist.

Why it matters: A 6x spread inside one job title means the market has not settled what it is buying. If you cannot write the job description yet, you cannot evaluate the hire either, and you may need the systems work done before you need the headcount.

Measurement, AEO, and AI search

A category got a name, and the ground moved anyway

Analyst

Forrester created a dedicated Answer Engine Optimization category, and named the test that comes next

Forrester, The Answer Engine Optimization Technologies Landscape, Q3 2026, published August 28, 2026. Surfaced via Profound, September 4.

Forrester has established a named landscape for AEO technologies, which is the mechanism by which a practice becomes a procurement category with its own evaluation criteria and vendor set. Profound, which is included in the landscape, quotes Forrester's forward view: the next frontier for AEO technologies is correlating KPIs like visibility and sentiment to business outcomes like sales, which will earn AEO bigger budgets.

Why it matters: Two things follow. First, AEO stops needing to be justified from scratch in a budget conversation, because an analyst firm has now defined it. Second, the honest measurement position we have held for months is now the analyst position too. Anyone selling AEO on visibility alone has roughly one planning cycle before the buyer asks the harder question, and the teams that have been careful about what can and cannot be attributed will be the ones with a defensible answer.

Original research

Company pricing pages are cited first in 12% of AI pricing answers, and Vendr and Reddit fill the gap

Kyle Poyar (Growth Unhinged) and Nikolas Laskaris (Profound), published September 3, 2026. 7,600 AI responses to pricing queries across the Cloud 100, six engines, collected July 29 to August 10, 2026. Profound is a Growth Unhinged partner.

Vendor-partnered studyCompany-owned pricing pages appeared somewhere in 46% of responses but appeared first in only 12%. No company in the set had its pricing page cited first in a majority of runs. ChatGPT was the friendliest, citing owned pricing pages first 38% of the time, while Google AI Mode, AI Overviews and Gemini typically buried them.

The three external domains that recurred: Vendr in 18.7% of runs, Reddit in 18.6%, and G2 in 15.9%. Of 77 public pricing pages examined, 57 were fully readable to bots; ten hid at least 40% of body content, usually through client-side rendering, interactive tabs and calculators, robots.txt blocks, or iframes. The study's named winners built what the authors call a pricing answer stack rather than a pricing page: Plaid's billing documentation was cited in 70% of its answers, ahead of its actual pricing page at 64% and FAQs at 50%.

Profound also documented fixing this on its own site. Its pricing was rendered client-side, so crawlers fetching raw HTML never saw the prices. Moving to server-side rendering, deployed June 25, was followed by a 13% week-over-week rise in citation bot traffic, and the pricing page becoming the second most cited page on the site. Vendor self-report, single case, no control.

Why it matters: This is the most immediately sellable finding of the week, because the diagnostic takes twenty minutes and does not require a tool. Fetch your own pricing page, or load it with JavaScript disabled, and see what a crawler actually gets. If the prices are not in the raw HTML, the model is going to describe your pricing using Reddit. For companies with no public pricing at all, the study found a vertical AI unicorn whose AI-reported price range spanned $100 to $2,400 per user per month, a 24x spread on one product. That range walks into the next negotiation.

Original research

ChatGPT Shopping switched retrieval sources in one day, and 517 tracked merchants felt it

Allen Wu, Profound Research, September 3, 2026. Analysis of 687 Profound customers tracking Shopping-mode prompts daily in July, narrowed to 517; roughly 1.75 million prompt runs in the daily July series.

Vendor claimProfound reports that feed-integrated retrieval overtook web search in ChatGPT Shopping on July 10, one day after OpenAI released GPT 5.6 models, rising from 8.26% to 61.54% in a day. Across its tracked base, 450 customers dipped 33% or more in Shopping visibility and 67 spiked 33% or more around the same date. A least-squares model jointly using owned web-search retrieval loss and feed-integrated gain explains 83% of observed visibility swings, per the appendix.

Merchant concentration increased with the shift. Top-10 merchant share went from 22.5% to 41.8% between July 7 to 9 and July 10 to 12, and unique merchants referenced fell more than 20%, from 13,524 to 10,607. Shopify-integrated retrieval accounts for roughly 35% of all feed-integrated retrieval and is described as the remaining route for smaller merchants without their own feeds. Profound quotes Shopify's own claim that AI searches powered by Shopify Catalog convert at twice the rate of those using scraped data, which is Shopify's number, not an independent one.

Why it matters: The methodology here is unusually transparent for vendor research, including sampling rates and correlation coefficients, which makes it worth citing rather than just reading. The strategic point is harder than the tactical one: a single model release reordered who is visible, with no warning and no content change. If your AI-visibility plan assumes content is the lever, know that the lever moved, and that some of it is now a feed integration decision with an owner in engineering.

Marketing and martech

The agent gets a job title and a seat

Product

Optimizely shipped agents named after roles, not features

Reported by Solutions Review, September 3, from Optimizely's announcement.

Optimizely launched Virtual Teammates, role-specific agents covering chief of staff, SEO and AI search analyst, marketing analyst, personalization strategist, and conversion-rate optimization manager. Each is configured with organizational permissions, connected systems, scheduled work, and retained context from prior projects.

Why it matters: Naming agents after roles rather than tasks is a positioning choice with an operational consequence. It forces the buyer to answer who the agent reports to, what it is allowed to touch, and what its work gets reviewed against. Those are the same three questions a real hire raises, and most marketing organizations have not answered them for software. A useful prompt for your own team whether or not you buy the product.

Product

Bazaarvoice, Phonely, Zoho and Capacity round out the week's releases

Compiled by Solutions Review, September 3, 2026.

Vendor claimsBazaarvoice introduced an AI Visibility package to prepare product content and user-generated content for discovery in AI-driven search and shopping, announced September 1. Phonely launched Alma, a language model built specifically for voice agents, trained on 10 million phone conversations with claimed sub-200-millisecond response times and claimed cost and speed advantages over general-purpose models. Zoho added agent-ready capabilities to its Catalyst platform, including Model Context Protocol support, Agent Skills, a non-interactive command line interface, and integrations with coding assistants. Capacity raised more than $54 million in Series E, bringing total funding above $159 million, and says it recently passed $100 million ARR.

Why it matters: Bazaarvoice is the one to note. A UGC platform packaging review content for AI retrieval is the same move as Profound's Shopping finding, arriving from the other direction: the sources a model reads about your product are increasingly being formatted for the model by the people who host them. If your reviews and ratings live on a platform, ask what that platform is doing to make them machine-readable, and whether you have any say in it.

Paid media and platform governance

Google tightened who is allowed to hold the token

Policy

Google's revised Developer Policies push integrations off proxies and onto their own Cloud projects

PPC News Feed, September 4, 2026, reporting Google's Ads Developer Blog announcement.

Google renamed the Google Ads API Policy to Developer Policies and revised it to eliminate reliance on programmatic proxies, requiring integrations to connect directly through a dedicated Google Cloud project. Google's stated reasons are unauthorized access risk and performance bottlenecks. The Ads API Compliance team is reviewing existing integrations and will contact affected developers directly.

Why it matters: Read this next to the write-capable ad protocols we covered last week. Every layer that lets one intermediary hold credentials for many advertisers is now under pressure, and Google's fix is to make identity traceable to a specific project rather than a shared pipe. If you or your agency run a homegrown or vendor-built ads integration, this is a real compliance item with a real review process attached, not a footnote. Find out this month whether anything you rely on sits behind a proxy.

Platform change

The AI Max auto-upgrade window is open right now, and Dynamic Search Ads are being retired into it

PPC Land and Search Engine Land on the migration timeline. Separately, PPC Land on language targeting removal.

Automatic migration of campaign-level broad match and automatically created assets campaigns to AI Max runs September 1 to 30, 2026, with in-account notices encouraging voluntary upgrade of Dynamic Search Ads to AI Max during the same window. Separately, Google is beginning removal of language targeting in September, roughly nine months past its original December 2025 target.

Why it matters: This is a live calendar item, not analysis. If you run broad match or DSA, you are being moved this month whether or not anyone on your side has read a release note. Two practical steps: check which campaigns are in the auto-upgrade set before September 30, and check whether any single-language campaign structure depends on language targeting that is going away.

The org chart and the budget

Money is moving, proof is the condition

Budgets

B2B marketing budgets are rising, and the benchmarks split hard by business model

MarketScale, September 6, 2026, citing Forrester, 10Fold, and The CMO Survey Spring 2026 via CMO.Works.

Third-party survey dataMarketScale reports Forrester finding 83% of B2B marketing decision makers expect higher investment over twelve months, and 10Fold finding 69% plan to raise budgets in 2026. The CMO Survey Spring 2026 benchmarks put marketing spend between 7.0% and 12.0% of revenue depending on model, with B2B product firms at 7.0%, B2C services at 7.2%, and B2B services at 10.1%.

In a companion item the same day, MarketScale reported BCG's finding that around 80% of CMOs have deployed generative AI for efficiency while marketing contracts stay under economic pressure, alongside Forrester's projection that 20% of new CMO job descriptions will ask for generative AI experience.

Why it matters: The B2B services benchmark at 10.1% of revenue is the number to keep handy, because it is the one most service businesses are unknowingly measured against. The tension in these two items is the whole 2027 planning story: budgets are going up and contract scrutiny is going up at the same time, which means renewals get decided on measurable cycle-time and productivity outcomes rather than roadmap.

Alignment

Influ2 puts a number on the marketing-to-sales handoff: 35%

MarketScale, September 6, 2026, on Influ2's alignment report, based on 105 Influ2 customers.

Vendor claimInflu2 sets a 35% hand-off rate from ad clickers to sales outreach as a benchmark, drawn from its own customer base of 105 companies. The report also cites Forrester data putting 63% of purchases at more than four people involved.

Why it matters: Small sample, vendor panel, so treat the 35% as a conversation starter rather than a target. The framing is what travels: if marketing cannot tell whether sales actually contacted the same named people who showed intent, alignment is a meeting rather than a process. That question is answerable in any CRM in an afternoon, and the answer is usually uncomfortable.

Put it to work

What to do with this, this week

Check your pricing page the way a crawler does. Fetch it with JavaScript disabled and read what comes back. If the prices are not in the raw HTML, that is the finding, and the fix is server-side rendering. Profound's own before-and-after shows what that can look like, with the caveat that it is one self-reported case.
Ask your team the stalled-deal question. "When a deal stalls, how long does it take us to find out why?" Salesloft's data says only about a third can answer immediately. If you cannot, the fix is stage-change evidence and field governance, not more AI.
Search Reddit, Vendr and G2 for your company name plus pricing. Those three domains appeared in roughly 18.7%, 18.6% and 15.9% of AI pricing answers. G2 and Capterra are editable. Reddit needs outreach. Both are cheaper than losing a negotiation to a number nobody at the company wrote.
Check the AI Max auto-upgrade list before September 30. Broad match and automatically created assets campaigns are being migrated during this window, and Dynamic Search Ads are being pushed in the same direction. Also check whether any campaign structure leans on language targeting, which is being removed this month.
Find out whether any ads integration sits behind a proxy. Google's revised Developer Policies require direct connection through a dedicated Cloud project, and its compliance team is reviewing existing integrations now. Ask the vendor, in writing, and keep the answer.
If you sell physical product, put feed integration on the roadmap. Feed-integrated retrieval now accounts for the majority of ChatGPT Shopping product recommendations by Profound's count. For smaller merchants the practical route is through Shopify. Frame it as an engineering ticket with an owner, not a content initiative.
Retire "are we using AI" from your planning. Everyone says yes. Replace it with: name one workflow where an AI recommendation is logged, reviewed, and attached to a number. If the room goes quiet, that is where to start.

Every claim above carries a source link. Figures attributed to vendors, to agencies reporting their own campaign results, or to companies surveying their own category are their claims, not independently verified facts, and are labeled as such. Two items sit slightly outside the coverage window and are included because they surfaced inside it: Forrester's Answer Engine Optimization Technologies Landscape was published August 28 and covered September 4, and Google's Developer Policies announcement was posted to the Ads Developer Blog in late August and reported September 4. The GTME Pulse active-listing counts are inconsistent on the source page and are presented here as directional. Coverage window: August 31 to September 7, 2026. Compiled September 7, 2026.