Skip to content
SourcepathBook a Demo

How Sourcepath works

Sourcepath doesn't start with advice. It starts with what AI actually answered.

Sourcepath runs the buyer questions you track against ChatGPT, Claude, Gemini and Perplexity, records what each one answered and which pages it cited, and derives the work from there — so every action it recommends is still attached to the evidence behind it.

Book a Demo
An illustration of how one Sourcepath recommendation comes to exist. The tracked buyer question is "Which customer data platforms handle enterprise identity resolution best?". Furthest away sits the answer ChatGPT returned for it: "For enterprise identity resolution, most buyers shortlist a small number of platforms. Competitor One is usually in that comparison, and the practical differences tend to come down to how each one handles deterministic and probabilistic matching." In front of that are the pages that answer cited: competitor-one.com, analyst-brief.org, industry-review.com, data-standards.org, practitioner-forum.net. In front of those is the observation Sourcepath recorded from them: "ChatGPT cites competitor-one.com answering this question, without citing yourcompany.com's own site." Nearest, and the only object in full focus, is the action derived from that observation: "Improve your existing page for this question", resolving to yourcompany.com/blog/cdp-vs-data-warehouse, with 12 evidence-backed requirements Sourcepath determined that page needs — three of them shown (Comparison table, Probabilistic matching, Supporting statistics) and the rest implied. Fine lines connect the action back through every piece of evidence beneath it, and remain visible once it has resolved. The order is the point: the evidence exists first and the recommendation is derived from it. Nothing here claims that any provider preferred a page, that a citation was caused by a page's contents, or that taking the action would cause a citation. The environment shown is representative, not customer data.

What gets recorded

It isn't watching your website. It's watching your buyer questions.

You choose the questions your buyers actually ask. On every scan Sourcepath puts each one to ChatGPT, Claude, Gemini and Perplexity, and keeps what comes back — the full answer, and the pages that answer drew its sources from. The four rarely agree, and the disagreement is itself evidence.

An illustration of one Sourcepath prompt-analysis record. The tracked buyer question is "Which customer data platforms handle enterprise identity resolution best?", and it has been asked on 8 scans. Four providers answered it. ChatGPT does not name Your Company. Its answer begins: "For enterprise identity resolution, most buyers shortlist a small number of platforms. Competitor One is usually in that comparison, and the practical differences tend to come down to how each one handles deterministic and probabilistic matching." It drew on 5 sources: competitor-one.com, analyst-brief.org, industry-review.com, data-standards.org, practitioner-forum.net. Claude names Your Company. Its answer begins: "Your Company and Competitor One come up most often for enterprise identity resolution. Your Company is generally described as strong on real-time resolution, though several sources note the implementation effort involved." It drew on 3 sources: analyst-brief.org, practitioner-forum.net, vendor-compare.io. Gemini does not name Your Company. Its answer begins: "Enterprise identity resolution generally relies on deterministic matching, probabilistic matching, or some combination of the two. Most platforms in this category support both, so evaluation usually turns on scale and data governance." It drew on 3 sources: techdigest.io, analyst-brief.org, data-standards.org. Perplexity does not name Your Company. Its answer begins: "At enterprise scale this is usually assessed on match rate, latency, and how conflicting records across sources are reconciled. Vendor documentation and independent comparisons differ noticeably on the last of those." It drew on 4 sources: industry-review.com, data-standards.org, schema-registry.org, cdp-guide.net. 1 of 4 name the brand, and none of the four drew on the brand's own domain. Nothing here ranks the providers or their sources, and the order in which sources are listed is the order the provider returned them, not a measure of importance. The environment shown is representative, not customer data.
  • ChatGPTBrand absent5 sources

    For enterprise identity resolution, most buyers shortlist a small number of platforms. Competitor One is usually in that comparison, and the practical differences tend to come down to how each one handles deterministic and probabilistic matching.

    competitor-one.comanalyst-brief.orgindustry-review.comdata-standards.orgpractitioner-forum.net
  • ClaudeBrand mentioned3 sources

    Your Company and Competitor One come up most often for enterprise identity resolution. Your Company is generally described as strong on real-time resolution, though several sources note the implementation effort involved.

    analyst-brief.orgpractitioner-forum.netvendor-compare.io
  • GeminiBrand absent3 sources

    Enterprise identity resolution generally relies on deterministic matching, probabilistic matching, or some combination of the two. Most platforms in this category support both, so evaluation usually turns on scale and data governance.

    techdigest.ioanalyst-brief.orgdata-standards.org
  • PerplexityBrand absent4 sources

    At enterprise scale this is usually assessed on match rate, latency, and how conflicting records across sources are reconciled. Vendor documentation and independent comparisons differ noticeably on the last of those.

    industry-review.comdata-standards.orgschema-registry.orgcdp-guide.net

One evidence foundation

The same answer. Three different questions.

AI Search Intelligence, Competitive Intelligence and Brand & Influence aren't three separate collections of data. They ask different things of the same recorded answer — whether you appeared and were cited, who else was in the room, and how you were described. That is why Sourcepath can tell you more than one number.

An illustration of one recorded AI answer being read three different ways. The answer is Claude's: "Your Company and Competitor One come up most often for enterprise identity resolution. Your Company is generally described as strong on real-time resolution, though several sources note the implementation effort involved." Its sources were analyst-brief.org, practitioner-forum.net, vendor-compare.io. AI Search Intelligence asks: Did we appear, and were we cited? It reads the words "Your Company" and records: Named in the answer — but yourcompany.com isn't one of its sources. 3 sources, none of them yours. Competitive Intelligence asks: Who else was in the answer? It reads the words "Competitor One" and records: Competitor One is named in the same answer, on the same question. 1 tracked competitor named. Brand & Influence asks: How were we described? It reads the words "strong on real-time resolution, though several sources note the implementation effort involved" and records: Mixed: a strength and a reservation, in the same sentence. Strength: real-time resolution · Concern: implementation effort. This reading is produced by a model rather than by a fixed rule. These three readings happen against the same record and in no particular order; none of them feeds another, and none is a stage in a sequence. Nothing here claims a provider preferred anyone, that being named caused anything, or that any of it predicts an outcome. The environment shown is representative, not customer data.

What it won't conclude

Sourcepath finds more than it tells you. That's deliberate.

A pattern needs a minimum amount of evidence behind it before Sourcepath will call it a finding — enough analyzed pages, and enough of them agreeing. Below that bar the evidence is still recorded, but nothing is said about it. Sourcepath does use AI, in the places where a model is the right tool. It just never lets one decide what the evidence means.

An illustration of Sourcepath declining to draw a conclusion. Five patterns were found across 7 analyzed cited pages for one tracked buyer question. a published date was found on 7 of 7 analyzed cited pages, which clears the bar, so Sourcepath records: "7 of 7 analyzed cited pages answering this question include a published date.". an FAQ section was found on 5 of 7 analyzed cited pages, which clears the bar, so Sourcepath records: "5 of 7 analyzed cited pages answering this question include an FAQ section.". FAQ schema was found on 4 of 7 analyzed cited pages, which clears the bar, so Sourcepath records: "4 of 7 analyzed cited pages answering this question include FAQ schema.". structured answer blocks was found on 3 of 7 analyzed cited pages, which is below the bar, so Sourcepath records no finding about it. analyst mentions was found on 2 of 7 analyzed cited pages, which is below the bar, so Sourcepath records no finding about it. The bar is fixed and applies to every candidate equally: at least 3 analyzed cited pages, and more than half of them sharing the pattern. Evidence below that bar is kept but never presented as a finding, and is not shown as a weak or low-confidence one either. Nothing here claims that a pattern causes a citation, that any provider prefers it, or that adopting it would change an outcome. The environment shown is representative, not customer data.

What you work from

What survives becomes work. With the reasons still attached.

Findings that clear the bar become prioritised work, and each piece of work keeps the observations it came from. Before your team spends a day on something, they can see the tracked question behind it, the pages already answering it, and exactly which findings put it on the list.

An illustration of one Sourcepath opportunity. It is centred on the tracked buyer question "Which customer data platforms handle enterprise identity resolution best?", and it is open. The recommended action is to improve existing page, resolving to yourcompany.com/blog/cdp-vs-data-warehouse. It carries 3 supporting observations. Visibility: yourcompany.com is not a source in any of the 4 answers to this question. Supported by 4 of 4. Competitive: ChatGPT cites competitor-one.com answering this question, without citing Your Company's own site. Supported by 1 of 4. Cited pages: 5 of 7 analyzed cited pages answering this question include an FAQ section. Supported by 5 of 7. Beneath them, the four providers that answered this question are listed, and none of them cited the brand's own domain. Nothing here estimates or projects an outcome: the action states what to do and what it came from, never what it will achieve. The environment shown is representative, not customer data.

Evidence-constrained execution

The work is decided before anything gets written.

Sourcepath determined what this page needs from the evidence, not from a prompt. When you open one of those requirements, AI drafts against it — with the recommendation, the page itself, the cited pages and the evidence excerpts already in hand. You decide what to apply. Then Sourcepath re-reads the result and checks whether the requirement is actually met.

An illustration of Sourcepath assisting with one piece of determined work on the page yourcompany.com/blog/cdp-vs-data-warehouse. The requirement being implemented is "Add an FAQ section to the page", which exists because it was observed on 5 of 7 analyzed cited pages. Sourcepath determined that requirement from evidence before any drafting began. The original passage reads: "Identity resolution stitches records from different systems into one customer profile. Most platforms combine deterministic and probabilistic matching to do it." The proposed version keeps it and adds: "Common questions: How accurate is probabilistic matching? What happens when two profiles conflict? How long does an initial identity graph take to build?" Adds a labelled question-and-answer block, which is what the requirement asks for and what the analyzed cited pages carry. The draft was informed by Recommendation, Document, Winning Pages, Evidence Excerpts, Implementation Requirements, Company Context. Related Recommendations was not available for this proposal and is shown unticked rather than assumed. Nothing is published: the person reviewing decides whether to apply the proposal to their working copy. Sourcepath then re-reads the resulting content and checks it against the requirement — FAQ Section Labeled and Question-and-Answer Structure — and reports complete. Completion is observed by re-reading the content, never asserted. The environment shown is representative, not customer data.

One intelligence, many depths

The same intelligence, at the depth each person needs.

A practitioner needs the provider answers, the citations and the page signals. Someone doing the work needs to know what to fix and why. Leadership needs to know where things stand. None of them need a separate source of truth — they need the same intelligence at a different resolution.

An illustration of one body of AI-search intelligence carried at three resolutions. It is drawn as a single band whose detail coarsens from one end to the other rather than as three separate panels, because nothing is added or removed between them. At the most detailed resolution, SEO and AEO practitioners investigate: Provider answers, Citations, Cited domains, Page signals, Analyzed pages, Brand mentions, Competitor presence, Scan history, Extraction method. At the middle resolution, Marketers and content teams work: Opportunities, Why it exists, Requirements, Create or improve. At the most distilled resolution, Marketing leadership understand: Sourcepath Score. At the most distilled end the Sourcepath Score brings together Visibility, Citation coverage, Brand Perception, and nothing else. They are drawn in proportion to how much each contributes, largest first; the exact weighting is not published here. Nothing is recalculated for a different audience. Reporting re-expresses the intelligence that is already there. The score measures neither revenue, traffic, conversion, market share nor causal impact. The environment shown is representative, not customer data.

Nothing is recalculated for a different audience. Reporting re-expresses the intelligence that is already there.

Simplified, never disconnected

The short version is still attached to what AI actually answered.

Leadership sees a conclusion. Underneath it Sourcepath still holds the findings that produced it, the evidence kept with each one, and the answer it recorded. Any of it can be followed back.

An illustration of a distilled Sourcepath conclusion opening back into the material underneath it. At the top, a Sourcepath Score of 25 — a single distilled figure, with no weighting or methodology shown. Beneath it the category that carries most of it, Visibility: Your Company is absent from most of the AI answers Sourcepath tracks. Beneath it, the 4 observations that conclusion is made of — ChatGPT: absent, Claude: appeared, Gemini: absent, Perplexity: absent. Beneath those, the Claude observation and the evidence kept with it. Brand mentioned — Your Company's name appears in the text of Claude's recorded response. The evidence set is: The recorded response, Your Company, 3 cited sources. Beneath that, the recorded answer itself. Claude was asked "Which customer data platforms handle enterprise identity resolution best?" and answered: "Your Company and Competitor One come up most often for enterprise identity resolution. Your Company is generally described as strong on real-time resolution, though several sources note the implementation effort involved." It drew on 3 sources: analyst-brief.org, practitioner-forum.net, vendor-compare.io. The brand's name appearing in that text is the observation — a text match that was recorded, and nothing more is inferred from it. Sourcepath can show you what a conclusion is made of. It can't tell you why a model wrote what it wrote — and it doesn't claim to. Sourcepath has no access to a provider's internal reasoning and makes no claim about why any answer was written. The environment shown is representative, not customer data.

Sourcepath can show you what a conclusion is made of. It can't tell you why a model wrote what it wrote — and it doesn't claim to.

Questions

What buyers ask about the platform. How the five capabilities fit together.

What is an AI Search Intelligence platform?

It is a system for seeing what AI assistants actually answer about your market, and acting on it. Sourcepath runs the questions your buyers ask against the monitored engines on a schedule, stores every answer with the pages it cited, and keeps that record so a finding can always be traced back to the answer it came from.

What can Sourcepath monitor across AI search?

The questions you choose, asked of ChatGPT, Claude, Gemini and Perplexity. For each answer it records which organisations were named, which pages were cited, and how your own domain featured in either — separately per engine, because the four routinely disagree and a blended number would hide the disagreement worth acting on. See how that monitoring works.

How do the five capabilities work together?

They read one body of evidence from five angles. Monitoring records the answers and their sources; competitive intelligence reads who else was named; brand influence reads the outside pages behind the answers; Content Studio turns the gaps into specific page work; reporting reads the whole thing again over time. Nothing is re-derived in a second place, which is why the numbers agree with each other. See how evidence becomes content work.

Is Sourcepath an AEO or AI search optimization platform?

Those describe the work; AI Search Intelligence describes what Sourcepath supplies for it. Answer engine optimization is the practice of earning representation in AI answers, and you cannot do it well without knowing what the answers currently say and what they cite. Sourcepath is the evidence layer underneath that work rather than a checklist on top of it. See the SEO and AEO team view.

Who on a marketing team uses it?

SEO and AEO practitioners work the cited-page evidence; content teams work the create-or-revamp decisions; communications teams work the third-party sources; leadership reads the same evidence summarised without losing what it rests on. One record, read differently by each of them. See the communications team view.

Now see what Sourcepath finds about your brand.

Book a Demo