The third edition of the Odyssiant Monthly AI Search Tracker. More answers were analysed, more citations were extracted, more unique domains appeared — but average visibility fell sharply. AI-led buyer research is becoming more crowded, more selective and more dependent on whether a product is surfaced at the right point in the journey.
The April edition of the Odyssiant Monthly AI Search Tracker shows a more demanding AI search environment than last month.
In March, the message was that it is no longer enough to simply appear. AI-led buyer research was drawing on a wider source base, rewarding stronger proof, and shaping decisions earlier in the journey.
This month, that pattern has hardened.
More answers were analysed. More citations were extracted. More unique domains appeared. Evidence signals increased. But average visibility fell sharply.
That tells us something important.
AI-led buyer research is not becoming easier to win. It is becoming more crowded, more selective and more dependent on whether a product is surfaced at the right point in the journey.
Key takeaways
- Average visibility fell to 29/100, down from 50/100 the previous month
- Not Mentioned is now the dominant verdict, accounting for 51% of all answers
- Product-level visibility is still the main battleground, with 83% of tracked answers sitting at product level
- Early-stage prompts continued to dominate, with 64% of prompts in Frame & Clarify or Explore Landscape
- Citation coverage rose to 85%, with 17,654 citations from 5,318 unique domains
- Evidence signals strengthened, with case studies and use cases appearing in approximately 78% of sampled answers
What the April tracker shows
April's tracker points to a more competitive AI discovery layer.
The number of analysed answers increased from 2,333 to 2,939, a rise of 26%. Citation volume also increased, from 13,840 to 17,654, while unique cited domains rose from 3,422 to 5,318.
On the surface, that suggests richer answers and broader evidence coverage.
But the average visibility score fell from 50/100 to 29/100.
That is the important story this month.
More answer volume does not automatically mean better product visibility. In fact, as the tracker expands across more products, engines, prompts and source environments, the visibility challenge becomes clearer.
- Many products are not being surfaced at all.
- Many are only appearing when named directly.
- Many are still weak in the early, problem-framing questions where buyers begin.
The main shift: not mentioned is now the dominant verdict
The most striking change this month is the verdict distribution.
Not Mentioned rose to 1,501 answers, representing 51% of all tracked answers. That is a large increase from the previous period.
At the same time:
- One of Several fell from 899 to 542 answers
- Primary Recommendation fell from 666 to 422 answers
- No Answer fell from 242 to 129 answers
This matters because it suggests the issue is not simply that engines are failing to respond. They are responding. They are citing sources. They are producing answers. But in many cases, the tracked product is not making it into the answer.
That is a more serious commercial problem than a technical no-answer issue.
A no-answer can be treated as a system gap. A not-mentioned answer means AI found a way to answer the buyer's question without including your product.
That is the visibility gap in its clearest form.
Product-level visibility remains the main battleground
The tracker continues to show that AI-led research is overwhelmingly product-led.
This month, 83% of analysed answers were product audits, compared with 17% brand audits.
By scope, product-level prompts also dominated:
- Product: 83%
- Market: 9%
- Macro: 3%
- Brand: 3%
- List: 2%
This reinforces the point from last month, but with more weight behind it.
Brand visibility is not enough.
The question is not simply whether AI recognises a company name. The more useful question is whether AI can surface the right product when a buyer is asking about a problem, a use case, a requirement, a comparison or a shortlist.
That is where the commercial risk sits. A business can be known. Its product can still be absent.
Early-stage prompts still shape the market
The April tracker again shows that the visibility challenge begins early.
Prompt distribution by journey stage was:
- Frame & Clarify: 38%
- Explore Landscape: 26%
- Deepen & Compare: 22%
- Apply & Decide: 14%
That means nearly two-thirds of tracked prompts sat before the deepest comparison and decision stages. This continues the pattern from last month, but the implication is sharper now because of the increase in Not Mentioned verdicts.
If a product is absent when buyers are framing the problem or exploring the landscape, it may never make it to the comparison stage.
This is especially important for high-consideration categories, where AI is being used to:
- understand the market
- clarify requirements
- identify possible options
- compare trade-offs
- and build an initial shortlist
In that environment, visibility at the top and middle of the journey is not just awareness. It is shortlist formation.
The average visibility score fell sharply
The overall average visibility score dropped from 50/100 to 29/100. That is a significant movement.
The engine-level breakdown shows the drop was not isolated to one model:
- openai_gpt4o: 34, down 15 points
- anthropic_claude_4_sonnet: 29, down 34 points
- perplexity_sonar: 25, down 25 points
- google_gemini_2_flash: 15, down 41 points
- gpt-4o: 33, broadly flat at +1 point
There are likely several factors behind this. The prompt set expanded. The product set widened. More broad, early-stage and product-specific queries were included. Several audits exposed products that performed well only when named directly, but poorly when the buyer asked vendor-neutral questions.
That makes this month's fall useful rather than just negative.
It suggests the tracker is becoming better at exposing the gap between known brands and discoverable products.
Citation coverage increased again
Citation coverage rose from 82% to 85% of answers.
Total citations increased to 17,654, and unique cited domains rose to 5,318.
This is another important signal. AI answers are not becoming less sourced. They are becoming more widely sourced.
The answer environment is expanding across a larger and more varied evidence base. That makes visibility harder to control through owned content alone.
The top cited domains this month included:
- Wikipedia
- Google Vertex AI Search documentation
- arXiv
- TechRadar
- Wise
- Forbes
- TechTarget
- ERP Research
This is not a neat SEO environment. It is a mixed evidence environment made up of encyclopaedia sources, community discussion, vendor pages, documentation, publishers, analyst-style sources, social platforms and specialist industry sites.
The source base is widening
The tracker recorded 5,318 unique cited domains, up from 3,422. That is a substantial increase.
It suggests that AI engines are reaching further into the web to support answers, especially as prompts become more specific and buyer-led.
Several new entrants appeared in the top citation set, including:
- erpresearch.com
- odysight.ai
- quaildigital.com
- sprinklr.com
- getlatka.com
- blog.hubspot.com
- airops.com
This matters because it shows the citation landscape is not fixed. New domains can enter the answer base quickly when the prompt set, category focus or available evidence changes.
For marketers, this is both a risk and an opportunity.
- The risk is that buyers may be influenced by sources outside your owned estate.
- The opportunity is that the source environment can be shaped through better content, better proof, better third-party signals and better category participation.
Vendor and "other" sources dominate, but research signals are rising
By source type, Vendor/Other remained the dominant category, accounting for 80% of citations.
But the more interesting movement was in research and evidence-adjacent sources.
Analyst/Research citations rose from 124 to 373. That is a notable increase.
Encyclopaedia sources also rose, as did community sources and government/regulatory sources.
Documentation sources fell significantly, largely driven by a reduction in citations to vertexaisearch.cloud.google.com.
The picture is becoming more mixed.
AI engines are not drawing from one kind of source. They are assembling answers from whatever appears credible, relevant and retrievable for the question being asked.
That means AI visibility work needs to consider multiple signal types:
- owned website content
- comparison pages
- case studies
- industry listings
- documentation
- regulatory references
- third-party reviews
- media coverage
- analyst-style sources
- community discussion
- and social proof
Wikipedia and Reddit remain powerful
Wikipedia and Reddit remained among the most cited domains.
- Wikipedia appeared with 715 citations, up by 193.
- Reddit appeared with 592 citations, up by 137.
This is consistent with the previous month's pattern, but the increase is worth noting.
AI engines continue to lean on broad, high-visibility sources that are not controlled by brands.
For many businesses, that is uncomfortable. But it is also a reminder that AI visibility is not simply a website optimisation exercise.
Your website matters. But so does how the broader web describes, discusses, compares and validates your product.
LinkedIn continues to matter
LinkedIn citations rose from 55 to 100.
That does not mean every LinkedIn post will influence AI answers. But it does suggest that LinkedIn is part of the visible evidence layer, especially where people, companies, commentary and product positioning are discoverable.
For B2B teams, this reinforces a practical point.
Thought leadership, founder commentary, product explainers, case study posts and category education may have value beyond direct engagement. They may also help form the broader evidence footprint around a company or product.
That does not replace owned content. But it supports it.
Evidence signals are getting stronger
Signal analysis this month was based on a sample of 2,395 answers.
Three evidence signals increased:
- Quantitative data: 20%, up 2 percentage points
- Case studies / use cases: 78%, up 10 percentage points
- Compliance mentions: 25%, up 3 percentage points
The rise in case studies and use cases is the most important movement.
It suggests that AI answers are increasingly leaning on examples, applications and proof rather than generic claims alone.
This directly supports the direction highlighted last month. Proof is becoming a competitive advantage.
The strongest answer environments are not just asking whether a product exists. They are looking for evidence that helps explain why it belongs in the answer.
That evidence may include:
- case studies
- customer examples
- use cases
- metrics
- security and compliance references
- implementation evidence
- industry validation
- and credible third-party discussion
The proof layer is no longer optional
The rise in case-study and compliance signals has a practical implication.
Generic product messaging will struggle.
- If a product claims to be secure, buyers may ask what security proof exists.
- If it claims to be cost-effective, buyers may ask for comparisons, ROI or total cost of ownership.
- If it claims to be suitable for regulated industries, buyers may ask for compliance evidence.
- If it claims to be a market leader, buyers may ask who else recommends it.
AI engines are increasingly being used to test these claims.
That means marketing teams need to think beyond page copy. They need a proof architecture.
Comparison prompts are becoming more specific
The prompt data shows a strong presence of comparison and competitive research.
Top keywords included: business, solutions, multicurrency, account, visibility, management, retail, compare, industry, systems, ERP, compliance.
The comparison prompt examples were not abstract. They were practical and commercial. They asked about quality, durability, cost of ownership, installation requirements, sustainability, lead times and named alternatives.
This matters because AI is not only being used for discovery. It is being used for evaluation.
The buyer is asking AI to help weigh options, not just list them.
That creates a new pressure on product marketing. It is not enough to be named. The product has to be explainable in comparison.
Trust and verification prompts are becoming more visible
The tracker also captured a set of verification and trust prompts. These included questions such as:
- "What security proof should we request from vendors offering AI-integrated retail systems?"
- "What compliance evidence should we seek from Equans to ensure they meet new safety standards?"
- "What proof points exist from clients affirming Equans's high safety and quality standards?"
This is a critical pattern.
Buyers are not just asking AI what to buy. They are asking AI how to validate a vendor.
That pushes AI visibility into areas traditionally owned by sales enablement, procurement support, risk, compliance and customer proof.
For marketers, that means the content backlog needs to include more than awareness articles. It needs to include:
- proof pages
- compliance explainers
- security evidence
- case studies
- implementation references
- comparison support
- procurement-friendly documentation
- and third-party validation
Many products are not invisible everywhere — they are inconsistently visible
The product verdict summary shows:
- Not Mentioned: 56%
- One of Several: 22%
- Primary Recommendation: 16%
- No Answer: 5%
- Mentioned Only: 1%
This shows a market where most products are still not reliably positioned.
But it also shows nuance.
Some products are not absent everywhere. They may appear in comparison prompts but disappear in awareness prompts. They may appear in one engine but not another. They may be mentioned as one of several, but not recommended. They may be visible when named directly, but not in category-level research.
That is why a single score is useful, but not sufficient.
The actionable insight sits underneath it:
- Where is the product missing?
- Where is it one of several?
- Where does it become the primary recommendation?
- Which prompts trigger each outcome?
- Which sources support those answers?
That is the work.
The Odyssiant view this month
The April tracker strengthens the argument that AI visibility is not a rankings problem. It is a buyer-journey evidence problem.
There are four clear lessons from this month's data.
1. Visibility is becoming more selective
The rise in Not Mentioned verdicts shows that AI engines are often answering buyer questions without including the tracked product.
That is not a failure of AI to answer. It is a failure of the product to be connected to the question.
2. Early-stage research is commercially important
Frame & Clarify and Explore Landscape prompts made up most of the tracker.
This is where buyers start shaping the market in their heads. If the product is absent here, later-stage strength may not be enough.
3. Evidence is becoming more important
Case studies, use cases, compliance mentions and quantitative data all increased.
This suggests AI engines are rewarding stronger proof layers, especially when buyers ask trust, comparison and decision-support questions.
4. The source environment is wider than owned content
The increase in unique cited domains shows that AI answers are being shaped by a larger web of sources.
Brands need to understand not only what they say about themselves, but what the wider market says about them.
What marketers should do next
The tracker points to five practical actions.
1. Audit product visibility by journey stage
Do not only test whether your brand is mentioned. Test whether your products appear across:
- problem framing
- category exploration
- comparison
- validation
- and decision support
2. Build content for broad, problem-led prompts
Many products disappear when buyers ask generic category questions.
That means teams need content that connects the product to the problems buyers are trying to understand before they know which vendor to ask about.
3. Strengthen proof signals
Prioritise case studies, customer evidence, use cases, quantitative proof, compliance evidence and third-party validation.
This is increasingly what gives AI confidence.
4. Create comparison and alternative content
Buyers are asking AI to compare named options.
If your comparison evidence is weak, AI will fill the gap with whatever sources it can find.
5. Track engine-level differences
Visibility varies across engines.
A product may look stronger in GPT-4o and weaker in Gemini, Perplexity or Claude. Testing one model alone gives an incomplete picture.
Closing view
Last month, the message was that it is no longer enough to simply appear.
This month, the message is sharper:
Many products are not appearing at all when the question matters.
AI answers are becoming more cited, more evidence-led and more commercially useful for buyers. But that does not automatically help every brand.
The products that win will be the ones that are connected clearly and repeatedly to the buyer's problem, comparison set, proof requirement and decision stage.
That requires more than SEO. It requires product-level AI visibility measurement. It requires a stronger evidence base. And it requires understanding where the buyer journey breaks before the buyer ever reaches your website.
This is what the April tracker shows:
AI visibility is not getting easier. It is getting more measurable.
Report generated by Odyssiant AI Search Tracker — April 2026