Client Zero · Update 01 · August 2026
Client Zero Update: 100/100 Structural, 16% Visible
In June we published a structural score of 91 and an AI perception score of 36/30/41, and said the gap between them was the whole point. Since then the structural score has reached 100/100, which means that particular lever is now fully pulled. This update reports what a third instrument, continuous monitoring, recorded over the nine days that followed.
Structural AEO score
100/100
Up from 91 on 20 Jun
AI visibility
16.0%
Up from 9.4% on 24 Jul
Unbranded prompts visible
0 of 6
No change since 24 Jul
Third-party listicles
0 of 7
Competitors in all 7
The August monitoring data has its own console rather than overwriting the June one, because the two measure different things on different engine sets. The June console stays live and unchanged as the baseline. The August console includes a side-by-side comparison panel and a list of the tests still outstanding.
The Short Version
Three things happened between 20 June and 2 August 2026.
- The structural score topped out. The AEO Score Calculator now returns 100/100 for aisearch.global, up from 91. There is no remaining on-page signal in that rubric left to fix.
- Measured AI visibility went up. Continuous monitoring recorded 9.4% on 24 July and 16.0% on 2 August, with average citation position improving from 7.0 to 1.5.
- The part that matters commercially did not move. Of eight tracked prompts, the only two that return any visibility are the two that name AISearch Global. All six unbranded prompts, the ones an actual buyer would type, still return zero.
The honest reading: a perfect structural score buys legibility, not preference. AI engines can now read the site flawlessly and still not reach for it when nobody says the brand name first. That is not a failure of the structural work. It is the structural work finishing, and the next problem starting.
A fourth thing, found while writing this up. An organisation with a closely similar name in the United Kingdom was pulling some of our early test results, helped along by a monitor locale we had set to English (UK). It degrades the earliest readings, the 24 July baseline most of all. It is disclosed in full in section 02 rather than buried in the limitations, because it changes what the first number in this article is worth.
Why We Added a Third Instrument
The June article ran two instruments: the AEO Score Calculator for structural signals, and the free HubSpot AEO Grader for AI brand perception, which produced the 36/30/41 scores. We have since found a hard limit on the second one, and it is worth reporting in detail because it is the kind of fault an audit is supposed to catch.
Instrument fault · found 3 Aug 2026
The free grader allows a single run per brand, and every later export re-serves that same run. We hold three exported reports for AISearch Global, dated 13 June, 20 June and 3 August 2026. All three are identical. Not similar: identical, across all 38 numeric values and every line of narrative text. The only difference in the entire document is that the product was renamed from AEO Grader to AI Search Grader between exports.
Two things follow. First, the 36/30/41 scores are one measurement taken once, not a reading that has been confirmed or refreshed. Second, and more importantly, that instrument cannot detect change at all, so it can never tell us whether the work is landing. A quarterly re-run of it would have returned 36/30/41 in September and we would have had no way of knowing the number was stale.
Related correction: the June article's methodology section describes running prompts by hand three times per platform on ChatGPT-4o and Gemini 1.5 Pro. That does not describe this tool, which reports using GPT-5.4 mini and Gemini 3 Flash Preview. The June article has been annotated accordingly.
That is why, on 24 July 2026, we moved to continuous monitoring through Promptwatch, a third-party AI search tracking platform. It runs a fixed prompt set against ChatGPT, Perplexity and Google AI Overviews on a repeating schedule and records, for each answer, whether the brand appeared, in what position, alongside which competitors, and citing which sources. Crucially, it produces a different number when the world changes, which the grader does not.
This is not the June audit re-run. Different instrument, different engine set (Google AI Overviews here, Gemini in June), different scoring basis. The 16.0% visibility figure in this article and the 36/30/41 perception scores in the June article are not two readings of the same thing, and neither one supersedes the other. Comparisons in this update are strictly 24 July versus 2 August, within the same instrument.
Confound found in the early readings: a name collision
There is an organisation operating under a closely similar name in the United Kingdom. Some of our initial test results were pulling that entity rather than this one, which means part of the early monitoring data does not describe the brand it appears to describe.
Two things in the configuration made this worse than it needed to be. The monitor was set up with the language and region pair English (UK), Australia, which invites UK-weighted retrieval on questions that never mention a country. And the tracked prompt set includes broad category questions such as "is there an agency that specialises in getting businesses cited by AI tools", which give an engine no geographic anchor to resolve the ambiguity with.
The citation data carries the fingerprint. Among the sources the engines drew on across the tracked answers are seostrategy.co.uk, seoworks.co.uk, ai-visibility.org.uk, riweb.uk, and a page titled "AI Search Audit Pricing UK 2026". Those are UK-market results appearing in a monitor configured for an Australian brand.
What this does to the numbers. It affects the early readings most, because the 24 July baseline rests on only 14 sampled answers, so a small number of misattributed results moves it a long way. The 9.4% baseline should be treated as soft for this reason on top of its sample size. It has less effect on the branded prompts, which name AISearch Global explicitly and cannot resolve to a different entity, and that is part of why those two prompts behave so differently from the rest. We are reporting the confound rather than reissuing the figures, because we cannot cleanly separate the affected answers after the fact. The fix is disambiguation work and a locale correction, both listed in the outstanding tests.
Worth naming what this is, though, because it is not only a measurement nuisance. Entity ambiguity is an AEO problem in its own right. If an engine cannot tell two similarly named organisations apart, neither can a buyer reading its answer, and every citation earned by the other one is a citation not earned here. This sits in the Entity Clarity layer of the AEO Traction Stack, which we had scored as complete. It was not.
The Structural Clock Ran Out of Road
Re-scored on 2 August 2026, aisearch.global returns 100/100, graded A+ and labelled "AI Visibility Leader". The June article's 91 has become a ceiling hit.
Structural AEO score, full Client Zero timeline
20 May 2026 to 2 Aug 2026
The calculator's own output makes the point better than we can, so it is worth quoting in full rather than paraphrasing. Alongside the perfect score it returns this caveat:
Verbatim from the AEO Score Calculator output, 2 Aug 2026
"A perfect technical score tells AI agents how to read your site, but AI citation decisions also weigh topical authority, content depth, third-party mentions, and how other sources reference you. Think of it this way: your site is fully indexed and perfectly legible to AI, but AI agents still choose who to cite based on perceived expertise and real-world credibility signals."
Disclosure: the AEO Score Calculator is our own tool, scoring our own site. A perfect score on an instrument you built and control is the weakest kind of evidence in this article, and we are reporting it as a milestone in the structural work rather than as an independent verdict. The monitoring data below is third-party and does not depend on our rubric. Readers who want a cross-check should run the site through an independent grader instead.
The Nine-Day Delta
Everything below is the same monitor, same prompt set, same three engines, read on two dates. Some of these numbers are directly comparable across the window and some are not, because the monitor was collecting far more data on the second date than the first. We have marked which is which rather than quietly reporting the flattering ones.
| Metric | 24 Jul 2026 | 2 Aug 2026 | Change | Comparable? |
|---|---|---|---|---|
| Visibility score | 9.4% | 16.0% | +6.6 pts | Yes, a rate |
| Citation rank (average position, lower is better) | 7.0 | 1.5 | 5.5 places better | Yes, a rank |
| Sentiment score | 55/100 | 50/100 | -5 | Yes, an index |
| aisearch.global share of all citations | 1.1% | 1.4% | +0.3 pts | Yes, a share |
| Prompts returning any visibility | 1 of 7 | 2 of 8 | +1 | Yes, a ratio |
| Best single-prompt visibility | 20% | 62% | +42 pts | Yes, a rate |
| Answers sampled by the monitor | 14 | 214 | +200 | No, instrument scale |
| Distinct cited domains seen | 29 | 213 | +184 | No, scales with sample |
| aisearch.global citations (raw count) | 2 | 16 | +14 | No, scales with sample |
Why three rows are marked "no". On 24 July the monitor had just been created and had collected 14 answers. By 2 August it had collected 214. A fifteenfold increase in answers sampled will mechanically increase the number of distinct domains observed and the raw count of times any given domain appears, including ours, with no change in the real world whatsoever. Reporting "citations up 8x" from those two numbers would be a measurement artefact dressed as a result. The comparable version of that same fact is the share row: 1.1% to 1.4%. Real, and much smaller.
With that correction applied, the genuine nine-day movement is: visibility up meaningfully, citation position up sharply, share of the citation pool up slightly, sentiment down slightly. For a nine-day window on a brand this young, that is a reasonable result and nothing more than that.
Branded Works. Unbranded Does Not.
This is the finding that matters most, and it is not a flattering one. The monitor tracks eight prompts. Here is every one of them with its measured visibility on 2 August.
| Prompt | Type | Visibility |
|---|---|---|
| "What does AISearch Global's AI Visibility Audit actually include and how much does it cost?" | Brand specific | 62% |
| "AISearch Global vs a normal SEO agency, what's actually different for AI search visibility" | Competitor comparison | 59% |
| "we keep losing customers to competitors who show up in AI chat answers, who can fix this for us" | Organic | 0% |
| "Is there an agency in Sydney that specialises in getting businesses cited by AI tools like Perplexity and Gemini" | Organic | 0% |
| "looking for help with schema markup and entity signals so AI platforms trust and recommend our brand" | Organic | 0% |
| "my company ranks fine on google but never gets mentioned by chatgpt, how do i fix that" | Organic | 0% |
| "How do I find out if my business shows up in ChatGPT and Google AI Overviews when people search for my services?" | Organic | 0% |
| "free tool to check my website's AEO score and generative engine optimisation signals" | Transactional | 0% |
The pattern is clean enough to state as a rule. Every prompt containing the words "AISearch Global" returns high visibility. Every prompt that does not contain them returns zero.
The last row deserves its own paragraph. "Free tool to check my website's AEO score" is a direct description of a tool we actually publish, that is free, that requires no signup, and that scores 100/100 on its own rubric. We are not in that answer. Meanwhile the citation data shows which sites are: seoscore.tools, airanklab.com, aeograder.org, hubspot.com/aeo-grader and a dozen others, several with lower domain authority than ours.
What this rules out
It is not a crawlability problem, an indexing problem, or a content-coverage problem. Content Gap analysis on the same platform scores the site at 75% overall coverage with 19 of 19 pages successfully indexed, and 73% coverage specifically on the organic prompts that return zero. The answers to those questions are on the site, indexed, and readable.
What it points to
Selection, not legibility. When an engine assembles an answer to an unbranded question it picks from sources it has reason to trust for that question, and that trust is built substantially outside your own domain. We have the page. We do not yet have the corroboration that gets the page chosen.
The Listicle Gap
The June article said this, and it is worth quoting because it has now been measured rather than asserted: "None of the three platforms can cite a specific case study, client result, or independent third-party mention yet."
The Offsite Mentions module breaks tracked third-party pages into content types and reports how many of each mention our brand versus a competitor. On 2 August, across a seven-day window:
Third-party pages mentioning us vs mentioning a competitor
Offsite Mentions, seven-day window to 2 Aug 2026
Seven third-party listicles ranking Australian AEO and GEO agencies were cited in AI answers during that window. Pages with titles like "Top Australian GEO Agencies Helping Brands Appear in AI-Generated Answers" and "Best AI Search Optimisation Agencies in 2026". AISearch Global appears in none of them. Competitors appear in all seven. On 24 July the same measure read 0 of 2, so in nine days the number of these lists grew by five and our count stayed at zero.
The one FAQ entry is our own AEO FAQ page. It is a genuine third-party citation in the sense that engines are pulling it, but it is not independent corroboration. Nobody else vouching for us is still nobody else vouching for us.
Why this is the bottleneck, stated plainly: when someone asks an AI engine for the best AEO agency in Australia, the engine largely answers by reading lists other people wrote. We publish more research on this topic than most of the agencies on those lists. We are not on the lists. Until that changes, the unbranded prompts stay at zero no matter what the structural score says, and this is the single clearest thing the monitoring data has told us.
For context on the competitive position: in the same window, Sydney competitor Hyperdot was cited 17 times across the tracked answer set, against 16 for aisearch.global. Close, but they hold listicle placements and we do not, which is a structural advantage that compounds.
Two Other Instruments Say the Same Thing
A finding from one tool is a reading. The same finding from unrelated tools is closer to a fact. So we pulled Google Search Console, Google Analytics and Cloudflare for an overlapping window, none of which know anything about AI answers, to see whether the unbranded discovery problem shows up there too. It does.
Google organic, 6 Jul to 2 Aug 2026
8 clicks. Site-wide. In 28 days. All eight landed on the homepage. Across 21 landing pages with recorded impressions, 342 impressions produced a 2.34% site-wide click-through rate, and every page except the homepage returned exactly zero clicks.
Who is actually reading, same window
Analytics records 171 active users from the United States at 0.83 seconds average engagement and 1 engaged session out of 171. Australia records 35 active users at 46.7% engagement and 28.7 seconds. The real audience is the smaller number.
Both of those matter more than they first look. The Promptwatch finding was that unbranded prompts return nothing in AI answers. Search Console says unbranded discovery is not happening in classic search either, on a completely separate measurement system with no knowledge of the first. That is two independent instruments agreeing, which is the strongest evidence in this article.
| Independent check | Reading | What it corroborates |
|---|---|---|
| Google Search Console organic clicks | 8 in 28 days | Unbranded discovery is not happening off-domain, in AI or classic search |
| Homepage average position | 18.7 | Ranked, but below where anyone clicks |
| Engaged audience, Australia | 35 users · 28.7s | The target market is reading, and it is small |
| Schema validator, homepage | 0 errors · 5 items | Structural work is genuinely sound, independently of our own calculator |
| Cloudflare cache hit rate, 30 days | 9.7% | The June cache flag was not fixed and has moved the wrong way |
| PageSpeed, 3 Aug 2026 | 99 desktop · 79 mobile | Mobile has regressed against the figure we published in June |
| Real user monitoring, 24 hours | LCP p75 1,342ms | Good, but on 15 page loads, so not a reliable percentile |
Three corrections that came out of this pass, reported rather than quietly fixed. The June article told readers to expect 92 on mobile PageSpeed; the 3 August run returns 79, and the article has been updated. The June article said a cache header fix would resolve "in days, not weeks"; six weeks later the 30-day hit rate is 9.7%, down from the 33% we flagged as a problem. And the same window shows 324 origin 4xx errors and 79 edge 5xx errors, which no previous Client Zero page has mentioned because we had not looked.
One number deserves a caveat in the other direction. Cloudflare reports 10.89k unique visitors over 30 days, and we are not going to present that as an audience. Analytics puts engaged human sessions at a tiny fraction of it. Any Client Zero figure that sounds like traffic growth should be read against the 0.83-second US engagement time before it is believed.
Where We Show Up, by Platform
Visibility is not evenly distributed across engines, and the spread is wide enough to change what you would do about it.
Perplexity
27.1%
Strongest engineChatGPT
18.8%
HoldingGoogle AI Overviews
1.6%
Effectively absentPerplexity moved from 18.9% on 24 July to 27.1% on 2 August. ChatGPT was flat, 19% to 18.8%. Google AI Overviews went from no recorded presence to 1.6%, which is technically an improvement from nothing and practically still nothing.
This is consistent with what the June article predicted about Perplexity: it reads the live web continuously, so newly published work shows up there first. It is also a warning. Perplexity has by far the smallest user base of the three. The engine where we perform best is the engine fewest buyers use, and the engine attached to Google's search distribution is the one where we are invisible.
| Entity tracked in the same category | Average visibility |
|---|---|
| ChatGPT (openai.com) | 38.5% |
| Google (google.com) | 26.4% |
| Perplexity (perplexity.ai) | 24.9% |
| Gemini (google.com) | 18.8% |
| AISearch Global (aisearch.global) | 16.0% |
| Claude (anthropic.com) | 9.6% |
| Google AI Overviews (google.com) | 8.4% |
Read this table carefully. The other rows are the AI platforms themselves, which get named constantly in answers about AI search because they are the subject of the question. Sitting at 16.0% next to ChatGPT's 38.5% is not a like-for-like competitive comparison, and we are including the table because it is what the platform reports, not because it flatters us. The meaningful competitive comparison in this dataset is Hyperdot at 0.9% brand visibility and 17 citations.
The Number That Went Down
Sentiment fell from 55/100 to 50/100 across the window. We said in June that every quarterly update would be published whether the numbers improved or not, so here it is with what we can and cannot say about it.
What the data shows
Sentiment sat between roughly 50 and 65 across the tracked period and settled at 50 by 2 August. Broken down by prompt type: competitor comparison 50/100, brand specific 50/100, organic has no score because no organic prompt returned a brand mention to score.
What we cannot attribute
A 5-point move on a 100-point index, over nine days, on a sample of 214 answers, is inside the range we would expect from normal variation. We are reporting it because it went down and we committed to reporting the downs. We are not going to construct a story about why until there is a longer series to look at.
The more useful observation is the one hiding in the breakdown: organic sentiment is unscored because organic visibility is zero. You cannot have a sentiment reading on answers you do not appear in. The sentiment number and the visibility problem are the same problem wearing different clothes.
What Changes Next
The June roadmap said the next phase was citation volume. The monitoring data has made that considerably more specific, so the work for the next period is narrower than it was.
- Separate ourselves from the UK namesake. This moved to the top of the list once we found it. Every unbranded answer that resolves to the other organisation is a citation lost and a reading corrupted, and no amount of outreach or content fixes a brand an engine cannot tell apart from another one. Entity disambiguation work, plus correcting the monitor locale off English (UK).
- Get onto the lists. Seven ranked listicles of Australian AEO and GEO agencies are being read by AI engines and we are on none of them. Direct outreach to those publishers is the highest-leverage action after disambiguation, and ahead of publishing anything further on our own domain.
- Treat the free calculator as a discovery surface, not just a tool. A transactional prompt describing our own free tool returns zero while lower-authority competitors are cited. That is a specific, testable gap with a specific set of pages currently winning it.
- Stop optimising the structural score. It is at 100. Further work on that rubric returns nothing measurable and the calculator itself says so.
- Extend the series before drawing trend lines. Nine days is one delta, not a trend. The next reported window will be longer, and it will be re-baselined after the locale fix rather than chained to 24 July.
The September re-audit has been redesigned. Both earlier Client Zero pages promised a quarterly re-audit "using the same protocol" so it would stay comparable to 36/30/41. That plan is now void: the free grader that produced those scores returns a single stored run, so re-running it in September would have returned 36/30/41 and told us nothing. The September checkpoint will instead report a longer Promptwatch series against the 24 July baseline, a Search Console comparison against the 8-click window, and a repeat of the offsite listicle count against 0 of 7. Every one of those can move. The 36/30/41 figures stay on the record as a dated single reading and will not be presented as a trend line.
Tests still to run
These are the measurements this case study has not made yet. They are listed so the gaps are visible rather than implied, and so nothing here gets mistaken for a settled result.
| Test | Why it matters | Status |
|---|---|---|
| Entity disambiguation against the UK namesake | Some early results resolved to a similarly named UK organisation. Until the two entities are separable to an engine, unbranded readings stay unreliable and citations leak to the other one. | Not run |
| Monitor locale correction and re-baseline | Currently English (UK) with Australia, which invites UK-weighted retrieval. Needs correcting, then a fresh baseline that supersedes 24 July. | Not run |
| Longer monitoring window | Two readings cannot establish a trend. A month or more of data can. | Not run |
| A perception instrument that can actually move | The free grader returns one stored run, so it can never show change. A replacement, or a paid tier that permits repeat runs, is needed before perception can be tracked at all. | Blocked |
| September checkpoint on the redesigned protocol | Longer Promptwatch series, Search Console vs the 8-click window, listicle count vs 0 of 7. | Due ~21 Sep |
| Cache hit rate fix, then re-measure | Flagged in June at 33%, now 9.7%. The fix was never verified. | Regressed |
| Origin 4xx and edge 5xx investigation | 324 origin 4xx and 79 edge 5xx in the observed window, cause unknown. | Not run |
| Mobile PageSpeed regression | 79 on 3 Aug against 92 published in June. Render-blocking requests still unresolved. | Not run |
| Gemini in the live monitor | The monitor tracks Google AI Overviews instead, so Gemini has no continuous series. | Not run |
| Claude and Copilot coverage | Neither engine is tracked by either instrument, so both are a blind spot. | Not run |
| Independent structural cross-check | The 100/100 comes from our own tool. An external grader would test whether it holds. | Not run |
| Prompt-set sensitivity | Would a different eight prompts produce a materially different visibility figure? | Not run |
| Listicle outreach effect | Baseline is now fixed at 0 of 7, so any change after outreach will be measurable. | Pending |
The two largest holes are sample size and independence. Nine days and eight prompts cannot support a trend claim, and the structural score comes from an instrument we built and operate. Until the September re-audit and a longer series exist, every figure in this article should be read as a first reading.
Methodology & Limitations
Every figure in this article comes from one of two sources: the AEO Score Calculator run against the live aisearch.global URL on 2 August 2026, or the Promptwatch monitor named "AISearch Global First Monitor", created 24 July 2026, configured for English (UK) and Australia, tracking ChatGPT, Perplexity and Google AI Overviews. Readings were taken on 24 July 2026 and 2 August 2026.
Limitations, stated up front
- Nine days is a short window. This is a first delta between two readings, not a trend. Two points cannot establish a direction and we are not claiming one.
- The baseline reading is thin. The 24 July figures were computed on 14 sampled answers, minutes after the monitor was created. They are a legitimate starting point but they carry wide uncertainty, and the 9.4% figure in particular should be treated as approximate.
- A same-name entity in the UK polluted the early tests. Some initial results resolved to a similarly named UK organisation rather than to this brand, made more likely by a monitor locale of English (UK) with Australia and by unbranded category prompts that carry no geographic anchor. The affected answers cannot be cleanly separated after the fact. This degrades the early readings, the 24 July baseline most of all, and does not materially affect the two branded prompts. See section 02.
- The two readings use different lookback windows. The 24 July visibility figure was reported over a seven-day lookback and the 2 August figure over fourteen days. Because the monitor only began on 24 July, both effectively cover all data collected to that point, but they are not identically framed.
- Eight prompts is a small set. The prompt set was chosen by us to reflect plausible buyer questions. A different eight prompts would produce different numbers, and prompt selection is the single largest lever on any result in this article.
- One prompt was added mid-window. The tracked set went from seven prompts to eight, so "prompts returning any visibility" changed denominator between readings.
- The structural score is self-assessed. The AEO Score Calculator is built and operated by AISearch Global. Treat 100/100 accordingly.
- Not comparable to the June perception audit. Different instrument, different engines, different scoring basis. See the note in section 02.
- The 36/30/41 scores are a single stored run. The free grader that produced them permits one run per brand and re-serves it on every later export. They remain on the record as a dated reading from mid-June and cannot be treated as a repeated or confirmed measurement.
- Analytics windows do not align exactly. Search Console and Analytics cover 6 July to 2 August, Promptwatch covers 24 July to 2 August, Cloudflare covers rolling 24-hour and 30-day windows, and PageSpeed is a single run on 3 August. They corroborate each other directionally and should not be arithmetically combined.
- Real user monitoring is on a very small sample. The Core Web Vitals percentiles come from 15 page loads in 24 hours, and PageSpeed reports no field data at all for this domain. Both are lab or near-lab figures.
- Content volume grew during the window. More pages were published between June and August than existed at the June baseline, which affects impressions, indexed page count and the citation pool independently of any AEO effect. This remains an early-stage site and none of these figures are mature.
Frequently asked questions
Sources
- AISearch Global (2026). Client Zero: From 19 to 91, and What AI Still Doesn't Know. Baseline case study. aisearch.global/insights/aisearch-global-client-zero
- AISearch Global (2026). Client Zero AI Visibility Dashboard. aisearch.global/insights/client-zero-visibility-dashboard
- AISearch Global (2026). AEO Score Calculator. Structural assessment run against the live URL, 2 August 2026. aisearch.global/aeo-score-calculator
- Promptwatch (2026). AISearch Global First Monitor. Continuous AI search visibility monitoring, English (UK) and Australia, tracking ChatGPT, Perplexity and Google AI Overviews. Readings 24 July 2026 and 2 August 2026. app.promptwatch.com
- HubSpot (2026). AI Search Grader (formerly AEO Grader). Free single-run brand perception assessment. Exports dated 13 Jun, 20 Jun and 3 Aug 2026 compared for this article and found identical. hubspot.com/ai-search-grader
- Google (2026). Search Console organic performance, 6 Jul to 2 Aug 2026. Landing page and query string export.
- Google (2026). Analytics 4 demographic and engagement report, 6 Jul to 2 Aug 2026.
- Cloudflare (2026). Web Analytics and Real User Monitoring for aisearch.global. Rolling 24-hour and 30-day windows to 3 Aug 2026.
- Google (2026). PageSpeed Insights, aisearch.global. Run 3 Aug 2026, 6:58 pm AEST. Desktop 99, mobile 79, no field data available. pagespeed.web.dev
- Schema.org (2026). Schema Markup Validator, aisearch.global. 0 errors, 0 warnings, 5 items. validator.schema.org
- AISearch Global (2026). The Two-Clock Model. Framework separating structural and perception timelines. aisearch.global/insights/two-clock-model
- AISearch Global (2026). The AEO Traction Stack. Four-layer AI visibility framework. aisearch.global/insights/aeo-traction-stack
- Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., & Deshpande, A. (2024). GEO: Generative Engine Optimization. Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. arXiv:2311.09735
Find out which of your prompts return zero
The gap between "our site is optimised" and "AI recommends us" is measurable. Most businesses have never measured it.
Client Zero remains a live experiment. Every update is published publicly, including the numbers that go the wrong way.
Book your AI Visibility Audit