Illustrative sample — every brand, metric and finding is fictional.
Get yours — real, freeA real briefing on a fictional company
This is exactly what CompetLab delivers — the depth, the structure, the analysis. Only the data is invented. Everywhere it says YourBrand, picture your company.
Strategic briefing for YourBrand
Last edition put one decision to you: hold product velocity flat for a cycle and spend it on getting named somewhere other than your own site. That is not what happened, and the record is unambiguous in both directions. Five dated launches went out in the last ninety days — Weekly Planner v2 on August 1, the agent endpoint on August 9, a summer roll-up naming eight shipped items on August 24, Slack workflows on September 8 and AI stand-up summaries on September 15. Over the same stretch, the number of pages both AI engines read for this market that name you stayed at zero.
Watch items carried in, resolved. Claude named you on a check where every question came back answered — the first of the two engine triggers to fire, and Gemini has still named you in none of the 15 answers it gave across the five checks in this window — three per check, and not one of them. Your AI Visibility Score reads 5. Your agent endpoint is still not discoverable at the standard address, and neither is any rival's. No first-party nimbusly page attacking you by name has appeared — probed again this run, none found. And pellowmark's comparison page still carries the claim that you have no public API, eleven weeks after your SDK's fifth release in the window.
And two closed differently. nimbusly's job board now carries a Head of Content Marketing at $150,000–$180,000, scoped by its own text to original research and earned media — the only marketing requisition open anywhere on this roster. And tovaqa's move upmarket, an open question for two editions, is closed by its own pricing page: the entry seat went from $11 to $19 on September 9, and a SOC 2 badge appeared the same week.
- 2026-09-28The set of pages both AI engines read for this market widened from 12 hosts to 14 between the two checks in this window. On both checks the number of them naming you was zero — the count grew because the tracked core widened, not because you fell off anything.Source: AI Sources · hosts read by both engines · Sep 28
- 2026-09-23nimbusly rewrote its homepage, and its messaging score moved 88 → 79 with a full captured copy change underneath it — a real editorial decision rather than a capture wobble. "Ship 2× faster with AI planning" became "The work platform for modern teams." What that tells you about where nimbusly is going is in the standings below.Source: nimbusly.example homepage capture · Sep 23
- 2026-09-20nimbusly opened a Head of Content Marketing requisition at $150,000–$180,000 — scoped to original research, flagship research properties and earned media rather than content production. It is the only marketing role open on this roster, and it is buying the thing this edition says decides AI answers.
You are not in the core of this market, and after 72 answers you are not separable from the brands below it either — the one standing that can still be argued either way. This edition names the fourteen pages where it gets argued.
Three editions in, the shape of the reading has not moved: you are the most machine-legible vendor this scan can measure and the least-recommended brand on the tracked roster. What is new is that the reading has become addressable. Two months ago the prescription was "get named somewhere off your own site" — a direction, not a task. This edition hands you the addresses instead, and there are few enough of them to work through by hand inside one cycle.
The first question is whether the AI models put you in the core of this market. They do not — and they have not ruled you out either. Pooled across the five checks in this window — 72 answers from five engines — you were named in 11: a 15% presence on an 8–25% range. A brand is in the core when the floor of its range clears 25%, and yours is 8. It is in the tail when the ceiling falls under 10%, and yours reaches 25. So you are in neither: named often enough that this market has not ruled you out, not often enough to be separated from the brands around you. Fennelbox holds exactly your presence and exactly your range.
The brands that are in the core are not close. nimbusly sits at 68 of 72 answers (94%, 87–98%), Cadenwork at 61 (85%, 75–91%) and Myralo at 58 (81%, 70–88%). None of those three ranges comes near yours, which is what makes the distance to the top of this market settled evidence rather than a bad month. Across the trend window every company including you carries overlapping ranges, so nobody's presence has measurably risen or fallen — what can be said is where each stands today.
On the September 28 check alone the five models returned 15 answers and named you in two, at an AI Visibility Score of 5 against nimbusly's 76. The score is a 0–100 figure combining how often the models recommend you with how high they place you when they do; below the top five, position contributes nothing, which is why a brand named twice can still read 5. It is the arithmetic under the standing above, not a replacement for it — a score cannot tell you whether you are in the core, and that is the question you came with.
The reach side is finished, and it is not the problem. All six AI crawlers whose access we check — ChatGPT, Claude, Perplexity, Microsoft Copilot, Google AI Overviews and Gemini Apps — are permitted to fetch your whole site, with nothing in your robots.txt restricting any of them; your file goes further than most vendors' and grants access explicitly rather than by omission. And they use it: the AI Sources check shows an engine opening a page on yourbrand.example in 12 of the 30 answers pooled across its two checks. On the answer side, that same check returns a verdict of recommended nowhere — 0 of those 30 answers named you, 0 of 16 from Perplexity and 0 of 14 from Google AI Overviews. On the latest check alone, Perplexity opened a page of yours while answering 4 of your 8 buying questions, and named you in none.
The path runs through other people's pages, and its great virtue is that it is short and countable. Fourteen hosts are read by both engines across these eight buying questions, and none of them names you. Eight belong to publishers, communities and one review site you can approach directly — pickledger.example, opsquarterly.example, stackwicket.example, remotework.example, productivitycompass.example, softwareledger.example, wickerstack.example and opsfolk.example — and the other six are other named companies' own sites, which the check itself marks as not targets. Five pages that Perplexity read do name you, out of the 196 it read, and every one sits on a host only Perplexity reaches: a software directory listing a plan you retired, a rival's comparison page carrying a false claim, a rival's attack page, and two pages on a tools guide. Those five pages are the entirety of your independent presence in this market's AI answers.
- 1The platform says you have four documentation pages and calls it a critical gap. It is wrong. help.yourbrand.example is a live 312-page help centre, every page carrying a modification date, the freshest dated September 24. The content scan reads root-domain sitemaps only, and the same blind spot understates every vendor here — nimbusly's documentation reads as 9 pages against 1,140 real ones at its docs subdomain.Source: help.yourbrand.example sitemap · Sep 28
- 2The same scan reads zero comparison pages against you, and you publish eleven. They live under
/compare/, a slug the categoriser does not recognise; nimbusly's 38 live under/vs/and count. Eleven pages that answer "how do you compare" — one of your eight buying questions — are invisible to the tool and, more to the point, were not opened by either engine on any of the 30 answers. - 3You have a free plan, and you are named on neither of the two price-shaped buying questions. On "affordable" the engines name pellowmark, Myralo and nimbusly; on "free plan" they name nimbusly and pellowmark and then wander into adjacent tools. Three of the six vendors here publish a free plan. You are the only one of the three the engines never mention when asked for one.
- 4Your review corpus is slowing. Three of your 96 Capterra reviews are dated in the last twelve months, against 41 in the twelve before that, and the newest is dated Jun 1 — while nimbusly sits at 4.5 across 1,180 and Myralo at 4.4 across 620. Your 4.7 is the highest rating on the board and it is being fed the least.
- 5You lead the field on agent adoption, and nobody has claimed the obvious first. Your scan scores 71, the highest of the five vendors it could measure — and no vendor here, including the two that run real MCP servers, publishes a server card at the standard address agents check first. That check reads 0 for all five.
- 6Three sites publish about you, and every one of them is wrong somewhere. pellowmark's comparison page says you have no public API while your SDK shipped seven releases in ninety days; pimblewick's attack page quotes an uptime figure and a customer count neither of which appears anywhere you publish; and apptrellis.example lists a $9 plan you retired in April.
Make your own best-tools listicle argue for you
Perplexity and Google AI Overviews opened yourbrand.example/blog/best-work-management-tools-for-small-teams in six of the seven answers where they read your site this check — and every one of those answers went on to name nimbusly, Cadenwork, Myralo, tovaqa and Brellwork, and not you. It is the single most-retrieved page you own.
Pitch pickledger.example's work-management round-up — it names eleven companies, not you
pickledger.example/best-work-management-software is read by both engines across five of your eight buying questions and names nimbusly, Cadenwork, Myralo, tovaqa, pellowmark, Brellwork, hinglework, pimblewick and three tools from outside the category. The check's own instruction for that host reads: "Pitch an update to this page: it is read for your market and names other companies, not you."
Publish an agent card at your well-known path — nobody in this market has one
You run the only publicly addressable MCP server on this roster, verified by a live protocol call, with 14 documented tools — and the check that looks for a server card at the standard discovery address reads zero for you and for every rival. The card is a transcription of what your documentation already publishes: the address, the authentication model, the tool list.
- Agent adoption: Your agent-adoption scan scores 71, against nimbusly's 58, tovaqa's 44, Cadenwork's 39 and pellowmark's 31 — the five vendors the scan could complete — with full marks on access control and content readability.
- A working MCP server:
agent.yourbrand.exampleanswered a live protocol call with a proper OAuth-style refusal, distinct from the generic error a bogus path on the same host returns — a real, auth-gated server behind 14 documented tools, one of only two MCP servers found anywhere on this roster and the only one reachable at a public address. - Machine legibility: You are the only vendor here publishing both an llms.txt and an llms-full.txt that read cleanly. nimbusly's llms.txt reads but it publishes no full file; Cadenwork and pellowmark publish neither.
- Security and trust: Security grade A at 94, all four headers present, top of a roster where nimbusly reads B/78 and pellowmark F/22; on the 26-signal trust dashboard you read 52 out of 100, "comprehensive," the highest live read of the set.Source: Tech & Trust scan · Sep 26
- Customers who cannot find a fault: Your Capterra rating of 4.7 is the highest on the board, with ease of use at 4.9 and customer service at 4.8; 58 of your 96 reviewers record no drawback at all, and none sits below 4.0.
- The one thing an AI has said about you: Claude's description, from the answer that named you this check: "a lightweight planner for teams that want a weekly cadence without a project-management office; strongest where the whole team lives in one tool, less proven above a few hundred seats" — placed 7th, marked as a recommendation.
The engines read you, and then write about everybody else.
Perplexity and Google AI Overviews open a page on yourbrand.example while they answer your market's buying questions, and then name other companies. That is not an access problem and it is not a legibility problem — both of those are settled in your favour, and settling them changed nothing. It is an attribution problem, and it has a shape: an AI answer is assembled out of what other pages say, and nearly every page in this market's retrieval set — the publisher round-ups, the community threads, the review directories, and one page of your own — carries somebody else's name.
Which is why a third consecutive edition of clean technical readings sits beside the standing you have. Everything you own is finished. What decides the answer is written somewhere else, by someone else, about someone else.
nimbusly was recommended in 14 of the 15 answers this check against your 2, and holds 87–98% of the pooled window against your 8–25% — two ranges that never touch, which is what makes this a settled gap rather than a bad month.
What changed this edition is what it is selling. On September 23 its homepage stopped arguing a number: "Ship 2× faster with AI planning" became "The work platform for modern teams," and a tagline about AI planning became "Everything your team does, in one place." A vendor that stops quoting a speed figure and starts naming the category has stopped arguing about features. It is the posture of an incumbent, and it lines up with a Series B in April and a job board that now carries 21 open roles.
Its hiring says the same thing in a different register. The one marketing requisition open anywhere on this roster is nimbusly's Head of Content Marketing, at $150,000–$180,000, scoped by its own text to original research and earned media — the exact class of third-party page this market's AI answers are assembled from. It is recommended in 14 of 15 answers already and is buying more.
Why it lands on you specifically: nimbusly.example is itself one of the fourteen hosts both AI engines read for these buying questions, across five of the eight. Its own pages are part of what the models consult before they answer — and by the check's own rule, a competitor's own page is not a target you can pitch. You cannot get onto it. You can only get onto the eight that are open.
- Any of the 14 hosts both engines read starts naming you — the check's already naming count moves off 0.Your first independent foothold in the pages that decide these answers. That count read zero on both checks in this window.
- An engine that opened your own page goes on to name you in the answer text.Seven times this check an engine read a yourbrand.example page and named only other companies. The first flip is the first evidence the page-level work landed.
- Gemini or Google AI Overviews names you in any check.Neither has named you once across the five checks in this window — Gemini in its 15 answers, Google AI Overviews in its 12. Claude and Perplexity have, so the wall is not universal.
- Your pooled presence range clears fully above 30%, or falls fully below 10%.It runs 8–25% today and overlaps a wide band of untracked brands, so nothing about your position is separable from the crowd. Either boundary would separate it for the first time.
- A Capterra review dated after Jun 1 appears — or your G2 count crosses 260 while Capterra holds at 96.The first says your review intake restarted; the second says the two pools are being fed through different channels rather than by the same customers.
- Any vendor here — you first — publishes an MCP server card at the standard address agents check first.Whoever publishes it first becomes the only vendor in this category an agent can discover without a human wiring it in, and you would know inside one check whether that was you or somebody else.
- pellowmark.example changes or drops the "no public API" line on its comparison page about you.The cleanest single test of whether a correction pitch lands on the rival pages that currently carry your facts.
- cadenwork.example or brunvex.example publishes a read-only or end-of-life date for Cadenwork.A tracked competitor recommended in 12 of 15 answers gets a migration window with a date on it — and you learn how long the answers keep naming a product that is being retired.
The decision last edition was a cycle of off-property proof. The cycle went to product. That is the serialisation this fork inherits, and it is worth being blunt about the cost of repeating it: another cycle spent the same way leaves the answer layer being written entirely by people who do not work for you — three of whom already publish pages about you, none of them accurate.
So the choice is not between marketing and product, and it is not about belief. It is about who owns the next cycle's calendar, and whether the list gets a name against it or stays a document.
One person, one cycle, worked top to bottom from the list: eight hosts to pitch or join, your own listicle, and the agent card. What makes this the defensible fork is that the work is sized in emails and single pages rather than quarters, and none of it is blocked on anything you have not already built. It is also the only fork with a read-out: five cheap checks next edition tell you whether it landed, and none of them requires a judgement call.
Your reviewers say the product is not the problem, and this edition's evidence agrees with them. If you take this fork, take it deliberately: accept that the AI answers will keep being written from pages you do not touch, and stop reading a flat score as a product signal. The one thing that makes this fork defensible is the review corpus — new dated reviews are the cheapest way a shipping company earns third-party pages without doing outreach, and yours has slowed to three in a year.
- The AI Visibility figures come from 15 answers on the September 28 check and 72 answers pooled across the five checks in the current window. Every share is stated with the count it came from, and every presence figure carries its own range.
- Where two ranges overlap the brands are not ordered — that is why Cadenwork and Myralo, and Brellwork and pellowmark, are shown as ties, and why nothing here claims your position rose or fell: every company on the trend carries overlapping ranges across the window.
- Threat level on the standings table is set by one rule and nothing else: how often the models recommended that brand on this check, relative to how often they recommended you.
- AI Sources counts answers, never percentages — eight buying questions per engine is too small a set for a share to mean anything. The engines report the pages they retrieved while answering, never which of those they leaned on, and pages we could not read are listed rather than counted against you.
- Google AI Overviews names companies in prose and describes none of them, so it counts toward how often you are named and toward nothing else. Its answer count in this window is 12 against 15 for each of the other four models, because it joined the check series one check into the window.
- All six homepages captured cleanly this cycle, so every messaging score below is current. That was not true last edition, and the comparison should be read as a fresh baseline rather than a movement.
- Content counts are pages a site publishes in its sitemap, not pages a search engine has indexed — we have no index access. The documentation and comparison-page counts read root domains and standard slugs only, and understate you on both.
- Pricing was readable on four of the five competitors this cycle. Cadenwork publishes no product price of its own — the Brunvex Suite bundle is a suite figure, and no claim here treats it as Cadenwork's.
- Security and trust rows exist for all six vendors this run; the technology-stack scan completed for five — pellowmark's fetch failed on all three attempts, which is a fetch failure and not a finding about pellowmark.
- G2 and Trustpilot counts could not be read from the source — those sites refuse every route available to us, including for URLs we invented as controls — so their figures are directional. Capterra was read page by page.
- Two readings could not be settled and are carried as open: nimbusly announced an MCP server on September 6 that none of the standard discovery channels could find; and Cadenwork's fold-in has no date on any page we could reach.
- Your agent-adoption comparison rests on five vendors — Myralo's scan failed on all three attempts, so it has no score, which is not the same as a low one.
This one's invented. Yours would be real.
Point CompetLab at your domain and your competitors, and a briefing like this one is built for you — on live intelligence, within two hours, then refreshed every cycle.