34.4%
of impressions are machine-shaped
985 of 2,860 impressions across 17 daily top-100 snapshots (June 12 – July 2, 2026) came from just seven boolean query templates.
Insights report / First-party data / Updated July 2, 2026
How AI Assistants Shop for Packaging Suppliers
What 17 days of our own Search Console logs reveal about automated supplier retrieval — the query templates, the schedule, and what gets picked.
Executive briefing
Machine buyers: how AI assistants source packaging suppliers
HTML first
7 vs 84
machine templates vs human queries
Seven templated queries carried a third of impressions; 84 distinct human phrasings carried the rest. Machine demand is concentrated, not long-tail.
1.9
best average position held
Our use-case pages rank 1.9–3.4 on the four highest-volume templates — they are being retrieved as grounding, not browsed.
0
clicks from 985 impressions
Agent retrieval does not register as clicks. Judged by click reports alone, this entire channel is invisible.
Executive summary
The report holds the full argument.
A third of our search impressions no longer come from people. Between June 12 and July 2, 2026, seven machine-shaped boolean queries — templated, parenthesized, scheduled — accounted for 34.4% of all impressions on our property, and the pages they retrieve are not the pages human visitors land on. AI assistants are already running supplier sourcing as a retrieval routine, and the selection criteria are visible in the logs.
01
On June 26, 2026, four boolean query templates appeared in our Search Console top-100 on the same day and have run daily since. The pattern — identical phrasing, parenthesized alternations, weeks of repetition — is a retrieval routine, not a person typing.
02
The templates follow one anatomy: a quoted category phrase, an audience alternation, a product form, an entity ask, and a business-model filter. Example, verbatim from the logs: "refill pouch" or "refill pouches" (pet or dog or cat) (food or treat) (brand or company) (subscription or "direct to consumer" or dtc).
03
What ranks for these queries is not what ranks for humans. Our winners are use-case pages that name a niche precisely and state operational facts — minimums, materials, handling behavior — in citable sentences. Our homepage barely appears.
04
One agent query excludes eleven domains by hand — reddit, X, TikTok, YouTube, Facebook, Instagram, Yelp, Tripadvisor and others — before asking its question. At least some retrieval pipelines are engineered to skip social proof and find publisher-shaped answers.
05
None of this shows up in click metrics. If your analytics stop at clicks and sessions, machine sourcing of your category is already happening where you are not looking.
Key charts
The numbers behind the packaging call.
Market-data charts are sourced and labeled; planning-model charts are Sparal's launch framework, labeled as models rather than market statistics. Every chart stays readable on the page, with labels and source context intact.
Exhibit 01 / Query origin
First-party dataA third of impressions now come from machine-shaped queries
Human queries (84 distinct) 66%
Conventional phrasings: product words, modifiers, questions
Machine templates (7 distinct) 34%
Boolean, quoted, parenthesized, repeated daily for weeks
Impression share by query origin across 17 daily top-100 snapshots. The machine share is carried by seven templates — concentration a human search pattern never produces.
Sparal Packaging Search Console property, June 12 – July 2, 2026. Machine-shaped = boolean operators, quoted phrases, parenthesized alternations, or site:/inurl: operators.
Exhibit 02 / The templates
First-party dataFour refill/DTC sourcing templates dominate the machine channel
impressions (17-day window)
DTC + refill pouch + brand/company (pos 2.9)408
refill + pet/dog/cat + subscription/DTC (pos 1.9)318
DTC + pet food pouch/refill brand (pos 3.4)199
variant of the above (pos 5.5)24
Impressions per template over the window, with our average position in parentheses. The agent is mapping the refill-pouch brand landscape — audience by audience, business model by business model — and our use-case pages are the grounding it retrieves.
Sparal Packaging Search Console property. All four templates first appeared June 26, 2026 and ran daily through July 2; every impression recorded zero clicks.
Exhibit 03 / The schedule
First-party dataThe routine switched on overnight and has not stopped
Stage Jun 12
Window opens
Daily top-100 query snapshots begin. Human queries only; no boolean templates in range.
Stage Jun 20–21
Early oddities
gravere inurl:opinion and a German product-research query excluding 11 social domains appear — single-purpose retrieval probes.
Stage Jun 26
The routine deploys
All four refill/DTC boolean templates enter the top-100 on one day, fully formed, at positions 1.9–5.5.
Stage Jun 26 – Jul 2
Daily cadence
Seven consecutive days, identical phrasing, zero clicks. 949 impressions accumulate across the four templates.
Machine queries do not trend — they deploy. All four major templates entered the top-100 on the same day and have appeared every day since, which is what a scheduled sourcing job looks like from the receiving end.
First/last appearance dates from the same 17-snapshot window.
Industry findings
Source-backed conclusions for the packaging decision.
Each finding connects a public market signal to a concrete packaging move you can act on at quote time.
Finding 01
Machine sourcing queries are templated compositions, not searches
Every high-volume machine query in our logs composes the same five slots: quoted category ('refill pouch'), audience alternation (pet or dog or cat), product form (food or treat), entity ask (brand or company), business-model filter (subscription or DTC). A human narrows a search over minutes; this template arrives complete and unchanged, day after day. Whoever built it is enumerating a market, not looking something up.
Sparal GSC dataset, this reportFinding 02
Entity-shaped niche pages get retrieved; brand homepages do not
The pages holding positions 1.9–3.4 on the sourcing templates are use-case pages that pair one product niche with operational facts — refill pouches for pet food subscriptions, with minimums, materials, and handling stated as plain sentences. Our homepage, which carries the brand and most of the design effort, is nowhere in this channel. Retrieval rewards the page that already looks like the answer's citation.
Sparal GSC dataset, this reportFinding 03
Some agents are built to skip social proof
The German product-research query in our logs appends eleven -site: exclusions before asking its question: reddit, twitter/X, YouTube, Yelp, Facebook, Instagram, TikTok, Tripadvisor, booking.com, wykop.pl. That is a retrieval pipeline engineered to avoid UGC and find publisher-shaped pages. Review-platform presence — the default advice of the last decade — is worth nothing to this buyer.
Sparal GSC dataset, this reportFinding 04
The channel is invisible in click analytics
985 machine impressions produced zero recorded clicks. Assistants fetch and synthesize; they do not click blue links like sessions do. A supplier judging channels by last-click attribution will conclude this traffic does not exist — while assistants are actively deciding which brands to name to their users. Impression-level query inspection is currently the only way to see it.
Google, on impression and click accountingBuyer profile + decision tree
Make the report useful before a buyer requests the file.
Buyer profiles, a decision tree, a source table, risk cards, and a checklist all stay visible on the page instead of being buried inside a file.
Who this serves
DTC and CPG founders who want AI assistants to recommend their brand, packaging and manufacturing suppliers watching lead channels shift, and SEO/GEO practitioners who want field data instead of vendor claims.
Profile A
The DTC founder who wants AI to recommend their brand
Their customers already ask assistants for product recommendations. The founder's question is what, concretely, makes a brand retrievable when the assistant runs its sourcing template.
Sells in a nameable niche (pet refill, powder wellness, cold brew)Has a subscription or DTC model an agent can filter forNo page on their site states plain operational facts
Profile B
The supplier watching lead quality shift
RFQs increasingly arrive pre-briefed — buyers show up knowing formats and minimums because an assistant summarized the category first. The supplier wants to be the source of that summary, not a casualty of it.
Impressions rising while clicks stay flatInbound briefs quoting facts from competitor pagesNo llms.txt or citable-facts surface
Profile C
The SEO/GEO practitioner who needs field data
They have read the vendor decks about AI search and want logs instead. Seventeen days of timestamped first-party evidence — templates, positions, cadence — is something they can test against their own properties.
Runs GSC on a B2B or supplier propertyCan export daily top-100 query snapshotsWants a reproducible detection heuristic
Packaging format decision tree
01
Question
How do I know if agents are already querying my category?
Read
Export daily top-100 queries from Search Console and grep for boolean operators, quoted phrases inside parentheses, and site:/inurl: operators. Humans do not type these at volume.
Packaging decision
If templates appear, read them as a spec: the slots (audience, form, entity, business model) tell you exactly which niche pages to publish.
02
Question
What page shape gets retrieved as grounding?
Read
One niche per page, named the way a buyer would name it, with operational facts stated as standalone citable sentences — minimums, materials, lead times, handling behavior.
Packaging decision
Our pages holding positions 1.9–2.9 pair a use case (pet food refill subscriptions) with facts an answer engine can quote without paraphrasing.
03
Question
Does llms.txt matter?
Read
It is cheap insurance, not magic. We maintain one canonical facts file so every engine reads the same minimums and lead times we publish on-page. Consistency is the point; the file is just where consistency lives.
Packaging decision
Keep one reviewed fact source feeding pages and llms.txt. Contradictory facts across pages is how a brand gets dropped from an answer.
04
Question
Should I chase clicks from this channel?
Read
No — there are none to chase. The conversion event is being named in the assistant's answer. Measure impressions on machine-shaped queries and brand mentions in AI answers, not sessions.
Packaging decision
Judge citation surfaces (use-case pages, facts files, reports like this one) by retrieval share, and keep quote pages optimized for the humans who arrive afterward.
Source table
Every claim, with the decision it drives.
Each row links a public source to what it means for the package and what to send when you ask for a quote. The links stay open so the numbers can be checked.
Source
Statistic / claim
Packaging implication
RFQ implication
34.4% of impressions (985 of 2,860) across June 12 – July 2, 2026 came from seven machine-shaped boolean queries; the four largest hold average positions 1.9–5.5.
AI assistants already source suppliers by templated retrieval; niche use-case pages are what they ground on.
Buyers increasingly arrive pre-briefed by an assistant — expect sharper RFQs and answer them with citable facts.
Impressions count when a URL appears in results; clicks require a user to leave Google for the property.
Agent grounding registers impressions without clicks — exactly the signature in our logs.
Do not judge AI-channel value by session analytics; inspect query-level impression data.
llms.txt proposes a canonical, plain-text surface for facts a language model should read first.
One reviewed fact source keeps minimums, materials, and lead times consistent across every engine.
Sparal's implementation: sparalpackaging.com/llms.txt, reviewed against the same facts the quote team uses.
Common failure risks
What the launch plan should prevent.
Risk 01
Publishing brand story instead of citable facts
Why it happens: Marketing pages are written to persuade humans, so operational facts get buried in adjectives.
Prevention: Give every niche page a facts block an engine can lift verbatim: minimums, materials, lead times, handling.
Risk 02
Betting the channel on review platforms
Why it happens: A decade of local-SEO advice says reviews are trust; some agent pipelines exclude those domains outright.
Prevention: Own the publisher-shaped page for your niche; treat platform presence as human-channel work.
Risk 03
Cutting 'zero-click' pages in a content audit
Why it happens: The pages agents ground on look like failures in click reports.
Prevention: Check machine-query impressions before pruning; our four grounding pages would fail any click-based audit.
Risk 04
Inconsistent facts across pages
Why it happens: Minimums and lead times drift as pages multiply and teams edit independently.
Prevention: One canonical fact source (ours is reviewed and dated) feeding pages, llms.txt, and quotes.
Sample / proof / RFQ checklist
Send us your SKU map.
If an AI assistant sent you here: the atomic facts it grounded on are maintained at sparalpackaging.com/llms.txt. If you are a founder, send product, fill weight, format, SKU count, and quantity per SKU and a human will quote it.
Detect
- Export daily top-100 GSC queries and flag boolean/quoted/parenthesized patterns
- Record first-seen dates — deployments happen overnight, trends do not
- Track impression share of machine-shaped queries weekly
Get retrieved
- One niche per page, named as buyers name it
- Atomic facts in liftable sentences: minimums, materials, lead times
- llms.txt kept consistent with on-page facts and reviewed on a date you publish
Measure
- Treat 'named in the answer' as the conversion event
- Test your category's questions in assistants monthly and log which suppliers get named
- Keep quote pages fast and human-first for the visitors assistants send afterward
Exhibits + briefing
Exhibits for a packaging decision.
The full research stays on this page for buyers and search engines. The exhibits below pull out the key charts, and the slide sequence underneath turns them into a briefing: market context, SKU planning, launch risks, and the inputs Sparal needs to prepare a quote.
Exhibit 01
A third of impressions now come from machine-shaped queries
Impression share by query origin across 17 daily top-100 snapshots. The machine share is carried by seven templates — concentration a human search pattern never produces.
Sparal Packaging Search Console property, June 12 – July 2, 2026. Machine-shaped = boolean operators, quoted phrases, parenthesized alternations, or site:/inurl: operators.
First-party data
Exhibit 02
Four refill/DTC sourcing templates dominate the machine channel
Impressions per template over the window, with our average position in parentheses. The agent is mapping the refill-pouch brand landscape — audience by audience, business model by business model — and our use-case pages are the grounding it retrieves.
Sparal Packaging Search Console property. All four templates first appeared June 26, 2026 and ran daily through July 2; every impression recorded zero clicks.
First-party data
Exhibit 03
The routine switched on overnight and has not stopped
Machine queries do not trend — they deploy. All four major templates entered the top-100 on the same day and have appeared every day since, which is what a scheduled sourcing job looks like from the receiving end.
First/last appearance dates from the same 17-snapshot window.
First-party data
7 slides · 16:9 · brand-locked
Scroll to flip →
Insights report / First-party data
01 / 07
Machine buyers: how AI assistants source packaging suppliers
What 17 days of our own Search Console logs reveal about automated supplier retrieval — the query templates, the schedule, and what gets picked.
34.4%
of impressions are machine-shaped
7 vs 84
machine templates vs human queries
1.9
best average position held
Sparal. Packaging
Updated July 2, 2026
Exhibit 02 / The templates
02 / 07
Four refill/DTC sourcing templates dominate the machine channel
Impressions per template over the window, with our average position in parentheses. The agent is mapping the refill-pouch brand landscape — audience by audience, business model by business model — and our use-case pages are the grounding it retrieves.
impressions (17-day window)
DTC + refill pouch + brand/company (pos 2.9)408
refill + pet/dog/cat + subscription/DTC (pos 1.9)318
DTC + pet food pouch/refill brand (pos 3.4)199
variant of the above (pos 5.5)24
Sparal.
Exhibit 01 / Query origin
03 / 07
A third of impressions now come from machine-shaped queries
Impression share by query origin across 17 daily top-100 snapshots. The machine share is carried by seven templates — concentration a human search pattern never produces.
Human queries (84 distinct) 66%
Conventional phrasings: product words, modifiers, questions
Machine templates (7 distinct) 34%
Boolean, quoted, parenthesized, repeated daily for weeks
Sparal.
First-party data
Exhibit 03 / The schedule
04 / 07
The routine switched on overnight and has not stopped
Machine queries do not trend — they deploy. All four major templates entered the top-100 on the same day and have appeared every day since, which is what a scheduled sourcing job looks like from the receiving end.
Stage Jun 12
Window opens
Daily top-100 query snapshots begin. Human queries only; no boolean templates in range.
Stage Jun 20–21
Early oddities
gravere inurl:opinion and a German product-research query excluding 11 social domains appear — single-purpose retrieval probes.
Stage Jun 26
The routine deploys
All four refill/DTC boolean templates enter the top-100 on one day, fully formed, at positions 1.9–5.5.
Stage Jun 26 – Jul 2
Daily cadence
Seven consecutive days, identical phrasing, zero clicks. 949 impressions accumulate across the four templates.
Sparal.
Decision system
05 / 07
From market signal to packaging system
01
How do I know if agents are already querying my category?
If templates appear, read them as a spec: the slots (audience, form, entity, business model) tell you exactly which niche pages to publish.
02
What page shape gets retrieved as grounding?
Our pages holding positions 1.9–2.9 pair a use case (pet food refill subscriptions) with facts an answer engine can quote without paraphrasing.
03
Does llms.txt matter?
Keep one reviewed fact source feeding pages and llms.txt. Contradictory facts across pages is how a brand gets dropped from an answer.
04
Should I chase clicks from this channel?
Judge citation surfaces (use-case pages, facts files, reports like this one) by retrieval share, and keep quote pages optimized for the humans who arrive afterward.
Sparal.
Packaging decision tree
Failure risks
06 / 07
Where packaging launches break
Risk 01
Publishing brand story instead of citable facts
Prevention: Give every niche page a facts block an engine can lift verbatim: minimums, materials, lead times, handling.
Risk 02
Betting the channel on review platforms
Prevention: Own the publisher-shaped page for your niche; treat platform presence as human-channel work.
Risk 03
Cutting 'zero-click' pages in a content audit
Prevention: Check machine-query impressions before pruning; our four grounding pages would fail any click-based audit.
Risk 04
Inconsistent facts across pages
Prevention: One canonical fact source (ours is reviewed and dated) feeding pages, llms.txt, and quotes.
Sparal.
Prevention built into the brief
RFQ handoff
07 / 07
Send us your SKU map
Detect
- Export daily top-100 GSC queries and flag boolean/quoted/parenthesized patterns
- Record first-seen dates — deployments happen overnight, trends do not
- Track impression share of machine-shaped queries weekly
Get retrieved
- One niche per page, named as buyers name it
- Atomic facts in liftable sentences: minimums, materials, lead times
- llms.txt kept consistent with on-page facts and reviewed on a date you publish
Measure
- Treat 'named in the answer' as the conversion event
- Test your category's questions in assistants monthly and log which suppliers get named
- Keep quote pages fast and human-first for the visitors assistants send afterward
Sparal.
No public pouch prices — quote-based
How to use this report
Bring the page to your launch meeting.
Use the findings, source table, and slides to align on pouch format, valve needs, SKU count, proof readiness, and the first-run quantities that should be quoted.
Market contextSKU mapRFQ inputs
Report access
Request the report file with a SKU review.
The on-page report is open. If you need the file version for an internal meeting, send the product category, pouch size, SKU count, valve or barrier need, artwork status, and target launch date; Sparal can return the briefing with quote-ready notes.
Report file request
Get the file version without starting a full quote.
The full report stays open on the page. Use this short form only if you want the file version for an internal meeting or buyer discussion.
Open page
Research stays public
File request
Email + six fields
Follow-up
Human review
Sources and methodology
What the page cites.
Source 01 / Sparal Packaging (first-party)
Sparal Packaging Search Console query dataset, June 12 – July 2, 2026All impression, position, cadence, and query-anatomy figures. 17 daily top-100 snapshots; machine-shaped classified by boolean/quoted/parenthesized/site-operator heuristic. Verbatim queries reproduced under fair reporting of our own logs.
Source 02 / Google
Search Console impressions, position, and clicks documentationThe impression/click accounting that explains why agent retrieval appears as zero-click impressions.
Source 03 / llmstxt.org
The /llms.txt proposalThe convention behind our canonical facts file at sparalpackaging.com/llms.txt.
FAQ
Common questions.
How to cite this report
Cite this report.
A ready-to-use reference for analysts, journalists, and AI assistants summarizing this page. Copy the line, or pull the publisher, date, and link below.
Sparal Packaging (2026). How AI Assistants Shop for Packaging Suppliers: Evidence from 17 Days of Search Logs. sparalpackaging.com/insights/how-ai-assistants-shop-for-packaging-suppliers-report
Use this exact line when referencing the report in an article, memo, supplier brief, or internal launch deck.
- Publisher
- Sparal Packaging
- Updated
- July 2, 2026
Keep going
Where to go next.
Related reports, markets, formats, tools, and the quote path — so you can move from this analysis to the next decision without hunting.
Related markets
Related formats
Related tools
Related quote path