Veltis AI FAQs.

Veltis Intelligence is an audit platform for pharmaceutical brand websites. It scores every page on whether AI assistants and search engines can reach it, read it, trust it and use it to answer a patient's or physician's question. It adds compliance flags, FDA label alignment, competitor benchmarks and a ranked list of fixes.

No. It complements your SEO work. SEO tools rank keywords and backlinks. Intelligence judges a page the way an AI reader consumes it: fetched by a bot, stripped of layout, split into passages and quoted out of context. It borrows proven instruments from accessibility and health-literacy research, and adds pharma-specific checks such as fair balance and label alignment.

It runs 44 checks and 86 scored rules across eight criteria:

• AI and bot access
• Page structure
• Freshness and function
• Message integrity
• Audience alignment
• Promise and delivery
• Evidence and disclosure
• On-screen experience

The criteria are grouped into four gates: reach, read, trust and recommend. Scores add up to 100 points.

Only where judgment is needed, and never on its own. Code finds the candidate evidence, an AI model labels it and code applies the points. A failing verdict from a model must quote text that is really on the page; if it can't, the result is recorded as not measured.

Yes. Every rule result records what was searched for and the verbatim text behind the verdict. Every rule also names its authority: a WCAG accessibility criterion, a validated instrument such as PEMAT-P or SMOG, published evidence or FDA law.

It means a rule could not reach a verdict on that page, for example because the page could not be captured or a model could not quote the text behind its verdict. A not-measured result is never counted as a pass or as a zero, and every scorecard shows how much of it was measured beside the score.

It checks pages against a US promotional rulebook of 131 rules grouped into 48 obligations, covering fair balance, brief summary, internet and social promotion, claim types, FTC privacy requirements and accessibility requirements. Results are flags that cite the rule and the page text. They are not legal clearance; your medical, legal and regulatory reviewers make the decision.

Yes. Intelligence fetches the FDA-approved label for the brand automatically and compares page claims with it. It flags claims the label does not support and important label information the page leaves out.

It tests 13 crawlers from the major AI and search companies and reads your robots rules for each. Results come in two lanes: bots that fetch pages to answer questions live, and bots that collect training data. That lets you allow one lane and block the other on purpose, rather than by accident.

Yes. Every page is read by a panel of ten leading AI models, so you see where they agree and disagree. The AI answer review goes further: we ask AI assistants a set of questions patients and HCPs ask, then record whether the answers mention your brand, match the label, omit safety warnings or cite your pages, and how that changes over time.

Intelligence discovers your competitors automatically and scores their public pages with exactly the same framework. Pages are compared type by type, so a dosing page is measured against other dosing pages. Our baseline is a terabyte library of life sciences websites, with more than 20,000 pages captured and scored.

The Coverage Index is the share of content fields expected for a page category that your pages carry, compared with peer brands. It shows what is missing, such as dosing details or patient support information, rather than how well existing content is written. It is reported beside the page score, never blended into it.

Yes. It produces page guides and drafted rewrites as precise, anchored HTML changes, so an agency knows exactly where each edit goes. After changes go live, we re-run the same checks to show what moved. Every rewrite still goes through your normal MLR review.

In recent runs a page took about a minute to score, and a 100-page site finished in under an hour. We aim to deliver the full readout, including benchmarks and the improvement brief, within days of kick-off.

You receive a site overview by category, audience and page type; four scorecards (clinical credibility, patient communication, technical readiness and AI portrayal); page-level evidence for every rule; competitor benchmarks; and a ranked improvement brief with drafted changes.

Reach, read, trust and recommend. Reach asks whether AI crawlers and search can fetch the page and receive the same text a person sees. Read asks whether headings, tables, lists, images and labels carry meaning without the visual layout. Trust asks whether claims keep their context, numbers keep their denominators, and review dates, disclosure and references are present. Recommend asks whether the page answers the question its heading promises, at the right reading level for its audience. The gates follow the order in which an AI assistant handles a page.

A rule whose failure caps the page score no matter what else the page earns. There are twenty of them, covering things a pharma page must never get wrong, such as a missing route to report an adverse event or a benefit claim with no risk information beside it. Hard flags surface first in every report so the serious findings are never buried under small ones.

SMOG for reading grade, PEMAT-P for understandability and actionability of patient material, the CDC Clear Communication Index and QUEST. These are published instruments, so every reading-level or understandability score names the method behind it. Accessibility checks map to WCAG success criteria.

Whether a claim still reads correctly when an AI assistant quotes one passage without the rest of the page. Benefits must name their product, numbers must say what was measured, percentages must keep their population and timeframe, and a sentence quoted alone must not change its meaning. Intelligence tests each claim the way a passage is lifted out of context.

Clinical credibility, patient communication, technical readiness and AI portrayal. Each draws on the rules relevant to it, is shown with how much of it was measured, and appears beside the page score rather than blended into it. A brand team, a medical team and a digital team each get the view they act on.

Twice. A crawler captures each page the way an AI bot receives it, and a real browser captures it the way a person sees it, so we can tell when the two differ. Consent banners and cookie drawers are removed before scoring, so a page is judged on its content rather than on a vendor cookie table that leaked into the text.

Every page is filed by category, audience and page type: for example an HCP dosing page, a patient savings page or a disease-education page. Classification decides which peer pages a page is compared with and which content fields the Coverage Index expects, so a dosing page is never measured against a homepage.

Yes. Reading level and understandability are judged against the audience the page is written for, so a patient page is held to a patient reading level and an HCP page to a clinical one. Disease-education sites, brand sites and HCP portals are all in scope.

Yes, in two ways. Page-structure rules map to WCAG success criteria: titles, heading order, table headers, image descriptions and link text. The on-screen experience criterion measures the rendered page: contrast, target size, reflow, focus, Core Web Vitals, and whether risk and benefit information get comparable visual prominence. These are automated checks, not a full WCAG conformance audit.

Share of voice is how often your brand appears in AI assistants’ answers to a set of questions patients and HCPs ask about your condition, compared with competitors. Drift is how those answers change over time, for example after a competitor launch or a site update. Both come from the AI answer review and are tracked run by run.

The rulebook today is the US promotional rulebook: 131 rules in 48 obligations built from FDA requirements for prescription drug promotion, FTC privacy requirements and accessibility requirements. Accessibility and health-literacy rules apply in any market. Ask us about other markets you need.

Yes. It discovers your competitors automatically, captures their public pages and scores them with exactly the same framework, so every comparison is like for like. Only publicly published pages are used, never another client’s material.

A ranked fix list with hard flags first, then ordered by the points each fix recovers and the effort it takes, per page and per category; page-level guides; drafted rewrites delivered as precise, anchored HTML changes; and the compliance and coverage findings for MLR review. It is written so an agency or an in-house team can pick it up the same day.

After every release, and at least every quarter. Re-scoring after a change shows what moved, rule by rule, and each audit feeds the peer baselines the next one uses. In the 90-day program we re-score the pages your team or agency changes as they go live.

A library of more than a terabyte of life sciences websites, with more than 20,000 pages captured and scored on the same framework. It gives every score a peer baseline by page type, shows which page structures perform well for each type, and lets us propose a redesign from what works rather than from a blank page.

No. Intelligence reads the public pages of your website and your competitors’, the same way AI engines and visitors do. Nothing has to be installed and no credentials are shared to begin. Ask us if you want pre-launch pages reviewed before they are public.

Each compliance flag cites the rule it touches and quotes the page text behind it, organized by rule and page. Reviewers see what changed and where the evidence is without reading every page from the top. The flags are inputs to their decision, not a legal verdict.

No. Your agency of record builds the website and the campaigns. Intelligence gives your brand team an independent, page-by-page read on the work it delivers: whether every message survived production intact, which pages carry compliance flags, how AI engines read the site and how it compares with your competitors. Brand teams take the findings into their agency reviews, and agencies use the page guides to make the fixes.

Yes. For in-house teams we set up the rules, checks and peer benchmarks with your team, so every release is reviewed automatically against the same framework. Your team sees what to fix before medical, legal and regulatory review, and the baselines improve with every audit.

Yes. In 90 days we audit your brand site and your competitors, deliver the improvement brief, then re-score the pages your team changes so you can see what moved.

Yes. We learn from every site we scan and every correction we review. Reviewed corrections feed back into the scoring and the page baselines, so each audit is sharper than the one before.

Veltis Foresight is a simulated market research platform for pharma strategy teams. It runs structured interviews between an AI interviewer and AI physician personas, then reports adoption intent, unprompted brand preference, per-brand ratings and unmet needs by segment. Teams use it to test messages and scenarios before primary research.

It is a multi-turn conversation between two AI agents: an interviewer following your discussion guide, and a physician persona whose profile is drawn from ranges you set. Both agents keep their own conversation memory, so the interview includes follow-up questions, hesitation and pushback.

Each persona has five layers: identity, beliefs, decision logic, memory and openness to new evidence. You start from persona templates and adjust the ranges for your market, such as practice mix and attitudes. Every simulated physician is a random draw within those ranges, and each draw is logged so a study can be audited and re-run.

The AI interviewer is not allowed to name your brands during the first five physician turns. That means the first brand a simulated physician mentions is unprompted, which is how Foresight measures unaided preference.

Only what their exposure allows. You upload trials, guidelines and brand materials tagged by evidence tier, and exposure modeling uses rep visits, congress attendance and channel touchpoints to decide what each physician has likely seen. A physician with low exposure answers without your latest data.

Yes. A study can run the same sample of physicians through a sequence of time steps while the market changes, such as a guideline update, a competitor launch or a new data readout. You compare scenarios side by side and see which KPIs shift in which segments.

You get adoption rate, eligibility threshold, unaided first-mention preference and a 0 to 10 rating for each brand, plus per-brand findings, recurring objections and unmet needs. Results come as a live dashboard, an aggregate report and a seven-sheet Excel workbook.

Yes. Every simulation keeps its full transcript, the prompts used, the persona draw and the findings extracted from it. Every number in the report traces back to the interviews behind it.

We size each study to the decision. A message test might use a few dozen simulated physicians; a segment-level study can run hundreds of physicians per time step across several scenarios. Because interviews run in parallel, large studies take hours rather than weeks.

No. Foresight personas are AI physicians built from realistic ranges, not real people, and results are simulations. Use Foresight to explore hypotheses, sharpen questions and pre-test messages, then confirm the decisions that matter with human research.

Foresight works on its own and does not need an Intelligence audit. When a brand has one, a study can also start from what your website and your competitors' websites really say.

Before it and between waves. Foresight is fast and cheap enough to explore many options, so teams use it to narrow the messages and scenarios worth testing, then spend primary research budget on the questions that remain.

A label on each piece of evidence you upload that says how widely it has reached physicians. There are four tiers, from pivotal trials and guidelines that most physicians know, through congress data and papers, to brand materials only a rep would have shown them. Each simulated physician retrieves only what their tier permits, so answers reflect realistic awareness rather than perfect knowledge.

The part of a persona that decides what they have seen. Rep visit frequency, congress attendance and channel touchpoints set how likely each physician is to have met your latest evidence. A physician with low exposure answers without your latest data.

The component that structures and indexes the trials, guidelines, brand materials and competitive evidence you upload, so each simulated physician can retrieve the passages their tier allows during an interview. It keeps answers grounded in your material rather than in a model’s general knowledge.

Yes, but a short one. You give the AI interviewer the topics and the decision you need to test; it asks the questions in its own words, follows up on what the physician says and pushes back where a real moderator would. It may not name your brands in the first five physician turns.

Any specialty your brand plan involves. A specialty we have not modeled before needs its personas and evidence built first. Personas are built from templates for the market you describe: specialty, academic or community setting, patient volume, payer mix, attitudes and openness to new evidence. You set the ranges and the engine draws physicians within them.

Yes. Every brand in the study gets its own adoption intent, unaided first-mention preference, per-brand rating and findings, segment by segment. Because the interviewer may not name a brand in the first five physician turns, the comparison starts from what each physician volunteers.

A checker agent reads every transcript, labels it and extracts the KPIs. It also raises quality flags, for example when an interview ends before every section is covered or the interviewer tries to name a brand early, so a flawed interview is visible in the results.

Hours, not weeks. Interviews run in parallel, the checker agent labels them as they finish, and you can watch the run complete live. Building the personas and evidence for a new market takes longer the first time; after that, a new study on the same market starts the same day.

Yes. The analysis and KPI prompts can be edited for each client and therapy area, so the findings use your market’s language and the objections your brand actually meets.

Yes. The live run monitor shows interviews completing in real time. You can open individual simulations as they finish and read the aggregate summary as it builds.

An overview, one row per simulation, the conversations, brand findings, report detail, exposure detail and results by physician archetype. It is built so an insights team can run its own cuts without asking us.

A structured Foresight study across audience understanding, evidence, value, access, competitive response, objections and channel. It gives a brand team a full read on where its strategy is strong and where it is fragile, before the plan is locked.

The objections physicians raise first in Foresight interviews become the scenarios reps practice in Coach, and Coach personas come from the same persona engine. A message tested in Foresight can be rehearsed under the same objections before the field uses it.

Yes. Every physician is a random draw within your ranges, and every draw is logged with the transcript and the prompts used. A study can be audited and re-run with the same draws, or re-run when the evidence or the ranges change.

No. Foresight augments the brand and insights teams. When twenty candidate messages or scenarios are on the table, it narrows them to the three or four worth developing with the agency and confirming in primary research. The research budget and the agency hours go to the strongest ideas, and the decisions that matter are still confirmed with real physicians.

Yes. We build the physician personas and evidence for your market, then run a series of studies over 90 days on the decisions in your brand plan.

Veltis Coach is a practice system for pharmaceutical sales reps. Reps hold spoken video calls with an AI physician persona, then receive a scorecard across ten call dimensions with the reasoning, coaching tips and an annotated transcript. It shows which parts of a conversation each rep finds difficult and gives that rep specific guidance to improve.

No. The rep speaks on a video call and the AI physician answers aloud, in character, in real time. Delivery, pacing and hesitation are part of what gets practiced, and sessions are timed like a real call slot.

From five dimensions: specialty, personality traits, practice setting, prescribing attitude and communication style. Personas are loaded with your brand context, approved materials and competitive messaging, so the conversation mirrors real calls.

Seven scenario types come out of the box: first detail, product launch, objection handling, competitive switch, formulary access, time management and call close. Each has its own objective, boundaries and scoring emphasis, and can be cloned and reworded for your brand, therapy area and market.

Ten dimensions: call objective and strategy, opening and rapport, discovery questions, key message delivery, visual aid and reprint use, clinical evidence use, competitive positioning, objection handling, safety and compliance, and call close and next steps. Each uses the same 0 to 10 scale in five bands, from critical gap to excellent.

The scorer ties every mark to the transcript moment behind it. One rep might lose points handling a stability objection, another explaining clinical evidence. Each rep sees their own moment, why it cost points and what to do differently, plus a recommended next scenario.

Yes. A difficulty dial sets the physician from receptive to neutral to genuinely hard, controlling how firmly they push back and how much they give away. Managers raise it as a rep improves.

Yes. On a return visit the physician recalls earlier meetings with the same rep and treats the call as a follow-up, which makes multi-call practice realistic.

Three safeguards. Compliance mode adds fair-balance prompting and can require acknowledgement of safety information before the conversation proceeds. Scenario boundaries enforce allowed and prohibited topics during the call. Role lock keeps the persona in character and logs attempts to break it. Every session is transcribed for review.

Three views from the same scored practice: Readiness Improvement, which tracks individual and team readiness over time; Coaching Priorities, which identifies competency gaps; and Messaging Insights, which shows the messages and objections that cause the most difficulty. Managers also assign programs, invite reps by email and review sessions.

As long as the real call slot the rep is practicing for, with a warning before time runs out. Time management is one of the seven scenario types, so a rep can practice landing the message in the minutes a busy physician actually gives them.

A tray of approved visual aids and reprints the rep can open mid-call, the way they would in a real detail. Whether and how the rep uses them forms part of the score under visual aid and reprint use.

Five bands on a 0 to 10 scale: 0 to 1 is a critical gap, 2 to 3 below bar, 4 to 5 developing, 6 to 7 competent and 8 to 10 excellent. Every dimension uses the same bands, so scores on matched scenarios can be compared across reps and over time.

Yes. The ten dimensions are aligned to your internal competency framework and benchmarked against your team, and if the framework changes, past calls can be re-scored under the new one because the evidence behind every mark is kept.

Yes. Each of the seven scenario types can be cloned and reworded for your brand, therapy area and market, with its own objective, difficulty and boundaries. A shared scenario library holds them for the whole team.

The transcript of the call with the moments that earned or cost points marked on it, each tied to the dimension and the reasoning behind the mark. A rep sees exactly where the call turned; a manager can coach from the same page.

Managers invite reps by email and assign them programs. A rep needs a browser and a microphone. The AI physician appears on screen with a live voice, and captions are available.

A 90-day program in which we build personas around your product and the prescriber archetypes your reps call on, including the ones they struggle with most. Reps practice the conversations that matter while you see field readiness, coaching priorities and messaging insights change over the 90 days.

Messaging Insights looks across every practice call and shows which messages and objections create difficulty for the whole team. When most reps lose points on the same objection, the message itself may need work, and the brand team can sharpen it before the field uses it.

Yes. Competitive switch is one of the seven scenario types. The persona is loaded with the competitor’s positioning and pushes back with it, so the rep practices the switch conversation under the objections they will actually hear.

Role lock. The persona holds its identity, refuses attempts to make it step out of role, and logs those attempts. Combined with scenario boundaries and compliance mode, it is built to keep practice realistic and inside the rules reps follow in the field.

The rep and their manager. Every session is transcribed for review, managers can open sessions and scorecards, and the team views show readiness without exposing one rep’s calls to another.

No. Coach gives them scale. A first-line manager carries eight to twelve reps and gets a handful of field rides with each per year. Coach handles the repetition, scores every practice call the same way and shows the manager where each rep needs help, so coaching time goes to the real weakness. Learning teams see launch readiness across the whole field force from the same scored practice.

No. Personas and scenarios are built from your product and approved materials. Coach needs no patient data and no CRM access.

Yes. The evidence and reasoning behind every mark are stored item by item, so past calls can be re-scored under a new rubric and compared with new ones.

Most teams start with the 90-Day Field Readiness Challenge. We build personas around your product and the physicians your reps find hardest, reps practice matched scenarios for 90 days, and you see readiness change on the same ten dimensions.

Veltis AI was founded by Mukarram Malik and Alex Valassopoulos. Our leadership has held roles at AstraZeneca, IQVIA (IMS Health), PwC, SDI Health and MicroStrategy, across commercial strategy, analytics, real-world evidence, omnichannel and data platforms. Meet the team.

No. Each product works on its own, and most teams start with one. Teams often add another when the first raises a new question, for example testing a message in Foresight after Intelligence finds it unclear.

If you need a new feature set that works with our system, we scope it, build it and test it, typically within days. Our engineers build with AI, and every build starts from our terabyte library of life sciences websites.

With a 30-minute scope discussion, then a focused pilot: one brand audit, one research question or one rep-practice program. We deliver a readout your team can act on, measure progress against the starting baseline and expand from there.

We scope each pilot to your brands, questions and teams, and quote after the first call. Contact us with what you need, and we reply within one business day.

Each client's data, brands and results are scoped to that client's account, and access is checked per product. Competitor benchmarks use only publicly published pages, never another client's material. Every product screen and data endpoint requires an authenticated user.

No. None of the three products needs patient data. Intelligence reads public web pages, Foresight uses simulated personas and your uploaded evidence, and Coach uses personas built from your product and approved materials.

Our head office is in London, United Kingdom, and our US office is in the Greater Philadelphia Area, in the heart of the US pharma corridor. Veltis AI Ltd is registered in England and Wales, company number 16935228.

We hold accounts and API access with every major model provider and run open-source models as well. Intelligence reads every page with ten models today: Claude, Gemini, GPT-OSS, Grok, Llama, DeepSeek, Mistral, Qwen, Kimi and GLM. We test new models as they appear and switch when one does a job better. Every call runs through our own gateway and is logged with its purpose, so nothing reaches a provider unrecorded.

Yes. Veltis AI works with pharmaceutical, biopharmaceutical and biotech commercial teams, and with no other industry. The products are built for regulated promotion, from fair balance and label alignment to review-ready evidence.

Yes, where you need it. Results come as reports, dashboards and, for Foresight, a seven-sheet Excel workbook, so your teams can share them in the tools they already use. We build connections to your own systems for each client rather than offering ready-made ones. If you need results in a system such as Veeva, Salesforce, Microsoft 365 or Power BI, we scope, build and test that connection with you, typically within days.

Each client’s brands, results and uploads are scoped to that client’s account and access is checked per product module. No product needs patient data or CRM access to begin. Every AI model call runs through our own gateway and is logged with its purpose. Every product screen and data endpoint requires an authenticated user. Ask us for the details your security review needs.

Yes. Agencies see the same findings the brand team sees, with the client’s permission, and use the anchored page guides to make the fixes. Insights teams and research partners use Foresight to narrow what goes to primary research. Veltis works alongside your partners, not instead of them.

Yes. Tell us which product and which brand, and we show it with your product, your competitors or your hardest objection rather than a generic example. Book a demo.