Results,
and everything we cannot prove yet

What did we measure on our own site?

On 6 August 2026 we fetched every route type on our parent site, outboundpros.io, using the documented GPTBot user agent, and recorded what a crawler that does not execute JavaScript received. The column heading carries the date because this is a dated observation, not a permanent property of the site.

What this measures. The size and content of the first HTTP response body served to a non rendering crawler, per route type, on one date. Method: one curl request per route with the documented GPTBot user agent string, byte count from the raw response body, no JavaScript execution, no cache warming. The empty baseline for this site is approximately 5,179 bytes, so any route within a few hundred bytes of that figure contained essentially no body content.

What this does not measure. It does not measure whether any AI assistant cited those pages, before or after. It does not describe what Googlebot sees, which is different, because Googlebot renders JavaScript on a deferred second pass. It is one site on one date and it is not a sample of anything. It says nothing about traffic, rankings or revenue.

Why we published the failing rows. Because the failing rows are the finding. A partially prerendered site is the common case and the hard one to detect, since a spot check will usually land on a route that passes. Publishing only the healthy rows would have produced a table that proved nothing. The parent's prerender coverage is scheduled to be extended to the homepage and the service and tool routes. When that happens, both the before and the after rows stay on this page, because the delta is the more useful dataset and it is the one nobody else can publish.

What is the scale of the business behind this site?

InboundPros is operated by the same team as a working B2B outbound agency. The table below is the one place in this group where the full record sits, so a figure quoted from anywhere else on this site traces back here. These are group facts, not results of any AI search engagement, and we are separating them for that reason.

What these measure. Scale and continuity of the operating business behind this site. A positive reply means a reply a human reviewer classified as interested, counted cumulatively across the group, not per client or per month. Infrastructure counts move week to week and are point in time snapshots rather than averages.

What these do not measure. They are not AI search results. They are not client outcomes for this practice. They say nothing about how any individual client performed, and a client count measures scale while telling you nothing about quality. We include them because the operating machine is the reason we have opinions about denominators, and because you should know who you are reading. One row that would normally sit here, a count of queued programmatic drafts, is left off entirely: it has no source file and no as of date, and an unsourced number inside a table row is the most extractable object on a page.

Where do the group's campaign numbers live?

Each of them lives on exactly one property, and none of them lives here. A verified figure is the most citable asset a group like this owns, and restating the same figure on four domains turns one strong source into four weak ones and hands an answer engine four places to disagree with itself. So the rule is one home per number, cited elsewhere as a clause with a link and never as a full restatement.

Three of those four properties are not live yet, which is why this section names them rather than linking them. The links get added as each site publishes, and the backfill is a dated step on the launch runbook rather than an intention.

What results do we refuse to publish?

Five things that could be on this page and are not, each absence deliberate and each one a thing our competitors publish.

No client case studies for AI search work. This practice is new. We have no engagement finished long enough to produce a before and after that would survive scrutiny. The honest thing to do with an empty results section is to leave it empty and say so, rather than to fill it with process descriptions dressed as outcomes.

No citation counts attributed to our work. Citation change is slow, lumpy, and confounded by model updates. A citation count without a frozen query panel, a stated model version and at least two quarters of the same method behind it is not evidence of anything.

No revenue or pipeline figures attributed to AI search. Attribution from an assistant answer to a closed deal is not currently solvable. Native app traffic sends no referrer header, so a meaningful share of the sessions involved arrive as direct traffic with no trail. Anyone showing you clean AI search revenue attribution is showing you a model, not a measurement.

No testimonials, star ratings, or review markup. No rating or review structured data appears anywhere on this site. Real testimonials will appear only when a named client has approved them in writing, and even then without rating markup.

No composite AI visibility score. The engines cite almost entirely different sources, so a single blended figure moves for reasons nobody can act on. We report per surface or we do not report.

What would change our mind?

Two findings would, and both are named here in advance so you can hold the page to them. A results page that cannot be falsified is marketing with a table on it.

If llms.txt were adopted as a production signal by any major provider, our position on it would change from ship it and expect nothing to something warmer. Currently adoption sits at 10.13% across roughly 300,000 domains in SE Ranking's May 2026 study, at zero among the top 1,000 domains by traffic, with no measurable citation effect once site authority, schema density and content recency were controlled for, and with no platform commitment from any model provider.

Method for this page. Every figure above traces to a dated source record and carries the date it was true. Nothing here is an estimate, a projection, or a range we chose. Where we wanted a number and did not have one, the sentence says so in plain words instead of carrying something plausible, and one table row was removed rather than filled.