Private to Netzer Precision and IV-Lead. Not for distribution.

The Answer Layer Audit · IV-Lead for Netzer Precision · 13 September 2026

Netzer is the answer. On two engines out of four.

Three months of measurement, 5,301 individual AI answers, re-read one at a time. Your dashboard averages those into a single 25%. Underneath it are four completely different companies.

HubSpot portal 8606367 · 21 tracked questions · 23 April to 16 July 2026 · 46,559 citations across 1,457 domains · read-only export

Findings

Seven things, ordered by how certain we are

Certainty first, size second. The first four are counted rather than estimated, and you can check every one of them in your own portal.

  1. 01

    You are the second most-named of the fifty vendors we counted, on 26.0% of 5,301 answers, behind HEIDENHAIN at 28.8% and just ahead of Renishaw at 24.8%. Second and third are 65 answers apart, so read this as “front rank”, not as a position.

    counted · 5,301 answers
  2. 02

    You are effectively absent from ChatGPT outside one territory. 83 of your 89 ChatGPT mentions come from the three space questions. On the other eighteen questions ChatGPT named you 6 times in 1,504 answers, which is 0.4%.

    counted · 1,753 answers
  3. 03

    One answer in three that reads your website never says your name. 1,945 answers cited netzerprecision.com and 627 of them did not name you. Roughly 350 of those named a competitor instead.

    counted · 1,945 answers
  4. 04

    You lost three quarters of your Perplexity visibility in five weeks and nobody saw it. 48.2% down to 9.5%, recovering only to about 25%. Measurement then stopped.

    counted · 13 weeks
  5. 05

    A third of your questions barely name any supplier at all. How often the engines name any company predicts your score, with a correlation of 0.87. Eight questions average 2.0%, ten average 46.2%.

    correlation · 21 questions
  6. 06

    Your tracked list accounts for 62% of competitor name-appearances, so 38% of the field is invisible to the dashboard. Five of sixteen slots hold vendors under 2%, one of which appears zero times in 5,301 answers, and a sixth holds an entry called “king kong”.

    counted · 50 vendors
  7. 07

    All 27 of HubSpot's recommendations have the same verb, which is create. None is categorised as fixing, reclaiming or retiring anything you already have, and two high-priority items target the single most unwinnable question on your panel.

    read from the portal
01

The average hides four different companies

Your dashboard reports one visibility number. It averages four engines that were not given equal numbers of runs, so it is not a fact about your market. It is a fact about how HubSpot allocated its budget.

Split it and the picture inverts. On Google's engines you are the most-named encoder company of the fifty we counted. On Perplexity you are third. On ChatGPT you are barely present, and outside one territory you are not present at all.

EngineAnswersYou were namedRateRankWho leads
Gemini Flash 3.560935057.5%#1Netzer Precision
Gemini Flash 2.51,16549642.6%#1Netzer Precision
Perplexity1,77444425.0%#3HEIDENHAIN, 31.7%
ChatGPT1,753895.1%~#10Renishaw, 26.4%
Who ChatGPT names, across its 1,753 answers to your 21 questions. Positions 7 to 10 are separated by 13 answers out of 1,753 and are inside sampling noise, so treat the bottom of this chart as a group rather than a ranking. What is not inside noise is the gap to the top: Renishaw and HEIDENHAIN are named five times as often as you are.

Why ChatGPT behaves differently, and why that is a separate problem

ChatGPT answered without searching the web at all in 842 of its 1,753 answers. Not truncated, not failed. Long, confident answers written from memory. In those 842 answers, Netzer was named zero times.

The other engines behave differently when they answer from memory. Perplexity names you in 19% of its unsearched answers, Gemini 3.5 in 25%. Their models know who you are. ChatGPT's does not.

EngineHow it answeredAnswersYou were namedRate
ChatGPTsearched the web911899.8%
ChatGPTanswered from memory84200.0%
Perplexityanswered from memory631219.0%
Gemini Flash 3.5answered from memory401025.0%
83 of 89
of your ChatGPT mentions come from the three space questions
6 / 1,504
times ChatGPT named you on the other eighteen questions
0.4%
which is your real ChatGPT rate outside space
0 / 842
times ChatGPT named you when it answered from memory

An absolute encoder reports its true position at power-on without searching for it. Half the time, so does ChatGPT. And it has no position for Netzer at all.

This separates two problems that look identical on a dashboard. Publishing better pages fixes the searched half. It does nothing for the unsearched half, which is a question of how widely you are written about on the sources these models are built from. Different budget, different people, different timescale.

02

Read, then not named

netzerprecision.com is the single most-cited domain in the entire dataset, with 2,680 citations, ahead of heidenhain.com and renishaw.com. The engines open your site more often than anyone else's.

And one time in three, they read it and never say your name.

1,945 answers cited netzerprecision.com. 627 of them, 32.2%, never said the name Netzer. Around 350 of those 627 named a competitor instead, and around 230 named no supplier at all, so roughly one answer in five reads your site and recommends someone else.

Perplexity · 11 July 2026 · "Which industrial automation encoders suit surgical robotics manufacturers?"

In its source list: netzerprecision.com/applications/surgical/

"Surgical robotics manufacturers prioritize compact absolute encoders… | Absolute Magnetic (e.g., AksIM-2) | Compact, off-axis design for space-constrained joints; robust against EMI from nearby motors. |"

Your surgical applications page was opened. AksIM-2 is an RLS product. Your page was in the source list. A competitor's part number was in the sentence.

EngineAnswers citing youCited and named youCited, named someone elseShare
Perplexity70141228941.2%
Gemini Flash 2.575149525634.1%
Gemini Flash 3.54123328019.4%
ChatGPT817922.5%
All four engines1,9451,31862732.2%

Being in the source list is not being in the sentence. Being named is not being recommended.

03

The question decides whether anyone is named. The page decides whether it is you.

We counted fifty vendor names against every one of the 5,301 answers: yours, your fifteen real tracked competitors, and thirty-four more that appear in your own citation data. That gives each question a number, which is how many company names the engines put in a typical answer to it.

It predicts your score almost perfectly.

Each dot is one of your 21 tracked questions. Horizontal: company names the engines put in a typical answer. Vertical: how often that answer named you. Pearson r = 0.869. Spearman rank correlation 0.882. Excluding Netzer's own mentions from the density count, Spearman falls only to 0.847.
4.85
names per answer · "Who produces space qualified rotary encoders?"
87%
your visibility on that question
0.14
names per answer · "What are the most common causes of encoder failure?"
0%
your visibility on that question

Across 251 answers to the failure-causes question, all fifty vendors together were named 35 times. Across 251 answers to the space question, 1,217 times. That is a 35-fold difference in whether the answer has room for a company at all.

Perplexity · "What are the most common causes of encoder failure in high-precision applications?"

"In high-precision applications, the most common causes of encoder failure are mechanical misalignment (specifically shaft run-out or a bent shaft), contaminationbearing wear due to vibration, and electrical noise interfering with signal integrity. … | Shaft Run-Out / Bent Shaft | The single most common cause… [6]."

Thorough, well sourced, and it contains no supplier at all. You score zero here because the answer has no vendor-shaped slot in it, not because your content is weak.

Question shapeQuestionsNames per answerYour mean scoreZeros
Selection: Which… / Who… / best… / top… / Compare…142.8336.2%0
Information: How do… / Why… / What causes… / What should…70.293.7%4

A third of your prompt set asks questions that barely name a supplier at all. Changing the verb is free and worth thirty points. Changing the page costs money and is worth ten.

04

You lost three quarters of Perplexity in five weeks, and nobody saw it

Because the reporting is blended and nobody was watching per engine, a very large event passed without comment. Between the first week of May and the first week of June 2026, your Perplexity visibility fell from 48.2% to 9.5%. It recovered only to about 25%. Your citations on Perplexity fell in step, which rules out a counting glitch. The engine genuinely stopped surfacing your pages.

Weekly. Blue: Perplexity. Grey: all four engines blended, which is the number on the dashboard. The blend never drops below 18.8% and hides a 39-point collapse. The dashed marker is the week HubSpot swapped Gemini Flash 2.5 for Flash 3.5, which lifts the blend for reasons that have nothing to do with Netzer.

The last run completed on 16 July 2026 and access had already ended on 15 July, two months ago. Nothing has been observed since. Whether the May collapse has returned is currently unknown, and there is no instrument running that would tell you.

05

Where the buyers are, and where you are not

Ten of your 21 questions are medical or surgical. Three slots are space, covering two distinct questions. The space questions are the ones you win, and they show what a fully earned answer looks like. Gemini Flash 3.5 named you in every one of 87 space answers. Medical is the bigger half of the panel and you are strong there on Gemini, mid on Perplexity, and absent on ChatGPT.

How often you were named, by territory and engine. The medical group is the ten questions written about medical and surgical applications; the space group is three question slots covering two distinct questions, one of which is tracked twice under different market tags. ChatGPT named you in 6 of 840 medical answers.

Perplexity · "Who produces space qualified rotary encoders?" · your best question, 87%

"| Netzer Precision | Develops contactless capacitive absolute rotary encoders (e.g., VLS family, Space GEO) tested per ESA/NASA standards for GEO, LEO, and deep-space missions [4][5][10]. |"

The engine did not describe you with adjectives. It named a product family, a product, and a qualification standard, because your space pages publish all three. That is the mechanism, and it already works on part of your site.

The contrast is the diagnosis. Your VLS-80 page carries TML and CVCM outgassing figures, TID and SEE radiation data, Parylene and Polyimide coatings, and three named standards. Your MRI and magnetic-immunity page says "Immunity to Magnetic". It sits at 3% across 251 answers.

06

Your citation list is the market map

1,457 domains were cited in answers to your questions. The top of that list is where your category's answers are actually assembled from, and several of the biggest entries are channels you have not claimed.

Most-cited domains across 46,559 citations. Blue: yours. Amber: an open channel that a competitor already occupies.

surgicalroboticstechnology.com was cited 569 times. Celera Motion holds a full company profile there, with six news items and a video. You have one news item and no profile. The site's own footer reads "Add Your Company". YouTube was cited 1,197 times.

The instrument itself is misconfigured

Every share-of-voice figure on your dashboard is measured against sixteen names someone typed in. Those sixteen account for 62% of all competitor name-appearances in your own data, so 38% of the competition comes from vendors the dashboard cannot see at all.

Tracked, but barely presentPresenceUntracked, but realPresence
POSITAL1.9%Gurley9.6%
Sensata1.5%NUMERIK JENA8.0%
TAMAGAWA SEIKI0.6%MACCON7.6%
Novotechnik0.1%Lika Electronic6.9%
BOGEN0.0%, zero in 5,301maxon4.5%
"king kong"junk entryMicronor1.7%

Five of sixteen slots hold vendors under 2%, and a sixth holds a junk entry. Gurley alone is named more often than four tracked names combined. Which means competitors mentioned = 0 has always meant "none of our sixteen", never "no vendor appeared". Location tags are split three ways as well, across US, United States and DE, with the same question tracked under two of them.

07

HubSpot's own 27 recommendations have exactly one verb

Twenty-seven recommendations are sitting in your portal, generated for 20 May to 19 June, all still marked NEW. Every single one carries the same action category: CREATE_CONTENT.

27 of 27
carry the action category "create content"
0
are categorised as fixing, reclaiming or retiring anything
14 / 13
split between your own site and LinkedIn, Reddit and YouTube
9.1%
best single estimated citation lift

Two of the high-priority items are "Create an educational article on encoder failure causes" and "Create a how-to guide on preventing encoder failures". Both target the one question on your panel where fifty vendors are named 35 times in 251 answers. The tool cannot see that, because it measures your absence and never measures the category's.

To be fair to the tool, it picked good surfaces. Thirteen of the 27 point at LinkedIn, Reddit and YouTube, which is the same direction as our own Tier 3 below. The problem is narrower than "it only tells you to blog", and it is not fixable by using the tool harder: it has one verb. It cannot say repair the FAQ markup that fails to parse, add product schema, rewrite the page sitting at 3%, claim the directory profile a competitor already holds, retire an unwinnable question, or fix the competitor list that distorts every number on its own dashboard. Those are six of our ten actions, and no setting in the tool will produce them.

The tool has one verb, and it is create. Six of our ten actions need a different one.

Plan

What to do, ordered by certainty rather than by size

Tier 1 costs nothing and we are certain of it. Tier 3 is the largest opportunity and the least certain. Do them in this order.

Free, certain, this week

Tier 1
  • 01Restore AEO measurementAccess ended 15 July, last run 16 July. A 39-point collapse happened inside the measured window and was never seen. Nothing has been measured since.Netzer with the HubSpot CSM
  • 02Fix the competitor listRemove "king kong" and BOGEN. Add Gurley, NUMERIK JENA, MACCON, Lika, maxon and Micronor. This is a measurement change, not housekeeping. Date it, and expect share of voice to fall once the real field is in it.IV-Lead, on your approval
  • 03Normalise the location tagsOne spelling per market. Today the same question is tracked under both US and DE, so any filter that groups by market silently splits.IV-Lead
  • 04Re-shape the seven information questionsKeep the product category and the market, change only the verb, from explaining to selecting. Deleting a prompt deletes its history, so Netzer performs every deletion, after the baseline is archived.Netzer deletes
    IV-Lead writes

Content and markup

Tier 2
  • 01Rewrite the magnetic-immunity page to the VLS-80 standardNumbers with units, part numbers, named standards, measured results. The spec is already written. Several fields need test data you may not have, which is a conversation rather than a copy task.Netzer supplies data
    IV-Lead writes
  • 02Repair the FAQ structured dataOne unescaped control character makes the whole block invalid, so search engines discard all of it. Add product schema with the specifications as properties while you are in there.Netzer web editor
  • 03Claim the surgical robotics directory profile569 citations. Celera has a full profile, you have one news item. The footer says "Add Your Company" and it is free.Netzer
  • 04Add sections, never rewrite a page that is already citedYour space pages are the most-cited assets in the category. Body rewrites have been measured cutting citation rates. Additions only.IV-Lead

The ChatGPT gap, which is a different problem and a different budget

Tier 3
  • 01Treat ChatGPT as a reputation objective, not a content objectiveZero mentions in 842 unsearched answers. Your own pages cannot reach that half at all. The target is the corpus these models retrieve and were trained on: trade press, engineering directories, standards bodies, encyclopaedic references, technical video, and the forums where engineers actually argue about encoder selection.IV-Lead with Netzer
  • 02Re-measure at 60 days on two dimensions, per engineNamed rate and cited-but-not-named rate, separately, for each engine. A blended number will hide the movement exactly as it hid the May collapse.IV-Lead
Limits

What this audit does not claim

  • No number, no date and no ranking is promised. We report what we find and what changed.
  • The mechanism is strongly supported, not proven. Pages with corroborable specifics get named, pages with adjectives get read and passed over. It is proven for you only by shipping a change and re-measuring.
  • Rates are computed on 5,301 answers, not 5,427. HubSpot's detail endpoint returns exactly six fewer answers per question than its own summary, on 21 questions out of 21. Coverage is 97.7%. We do not know what the six are.
  • Retiring zero-scoring questions raises the portal average arithmetically. Every report after that change has to carry the date, or it claims an improvement that did not happen.
  • Your name and your competitors' names were counted by two different methods. Yours uses HubSpot's own brand detector, with all its spelling variations. The other 49 use a single exact-word search each. Our independent count of "Netzer" matched HubSpot's detector exactly, at 1,379 both ways, which is reassuring but does not make the two methods identical. Treat small gaps between you and a neighbour as ties.
  • Company groups are counted as separate names. RLS and Renishaw, and Novanta and Celera Motion, are related businesses counted separately here. Group them and the ordering at the top of the table moves.
  • One question is tracked twice. "Which rotary encoders are most reliable for space and satellite systems?" appears under both a US and a DE market tag, so it carries double weight in every panel-wide average in this report, including the 26.0%.
  • There are three different citation totals and they measure different things. 46,559 is every source link the engines returned. 41,859 is HubSpot's own de-duplicated count over the same answers. Your dashboard shows 42,782, which is its count over all 5,427. The domain chart uses the first.
  • Vendor counts are a floor, not a census. Fifty names were counted. A vendor never cited and never tracked would be missed, and acronyms that are also ordinary words, including SICK, RLS, ifm and FLUX, were matched case-sensitively, so a lower-case mention is missed.
  • Answer engines are not deterministic. Any single answer a colleague happens to see means nothing. Judge on the rate, over hundreds of runs.
  • Nothing in your portal was changed. Read-only methods only. Settings were never opened and no CRM record was read.