Ahrefs October 2025 Updates: MCP, AI Citations, and What to Measure Next

Ahrefs shipped 16 product updates in its October 2025 roundup, led by a hosted MCP server connection for ChatGPT and deeper Brand Radar reporting for AI citations. For marketers, the useful takeaway from these Ahrefs October 2025 updates is not simply that more AI data is available: it is that we can build a more disciplined process for checking whether our brand appears, is cited, or loses the answer to a competitor on the buyer prompts that matter.

The release combined several different categories of work: connecting Ahrefs data to AI assistants, extending historical chatbot data, adding URL-level entity tracking, and improving conventional SEO, analytics, social, and API workflows. We should treat those as separate capabilities with separate measurement jobs—not as proof that an AI visibility problem has been solved. (ahrefs.com)

What changed in the Ahrefs October 2025 updates

The October 2025 Ahrefs release was published as a roundup of 16 updates. The most relevant changes for AI search measurement were the hosted Ahrefs MCP Server, Brand Radar’s expanded chatbot history, URL entity tracking, and a new citations chart in the Brand Radar Overview.

Here is the practical breakdown:

• Hosted Ahrefs MCP Server: Available to Lite plans and above, letting users connect supported AI tools to live Ahrefs data without hosting their own server.
• Historical chatbot data: Brand Radar added up to five months of history for ChatGPT and Perplexity, plus three months for Gemini and Microsoft Copilot.
• URL entity tracking: Teams can track an entity by name variations and URLs, rather than treating a brand name as the only signal.
• Citation reporting: Brand Radar added a citations chart alongside mentions, impressions, and AI Share of Voice.
• Prompt-reporting usability: AI Responses gained a chart view, Topics gained nested tables, and reports could retain filters, be named, and be duplicated.

Ahrefs also updated Web Analytics, Site Explorer, Google Search Console reporting, Social Media Manager, and its API. Those are valuable operational additions, but they do not all answer the same question. A Google Search Console SERP overlay helps investigate traditional rankings; a Brand Radar citation chart helps observe AI answer sourcing; an MCP connection helps query Ahrefs data conversationally. We should avoid rolling all three into one vague “AI performance” metric. (ahrefs.com)

The Ahrefs MCP Server: connecting Ahrefs to ChatGPT

The headline feature was the remote Ahrefs MCP Server. MCP, or Model Context Protocol, is a way for an AI assistant to use an authorized external data source or tool. In this case, Ahrefs hosts the server, so a user does not need to run infrastructure locally before connecting an Ahrefs account.

Ahrefs documents MCP access for accounts on Lite plans and above, and its current documentation describes connections for ChatGPT, Claude, Copilot, and other supported AI tools. The connection gives the assistant access to Ahrefs actions and data subject to plan limits, row limits, and API-unit allowances. (docs.ahrefs.com)

What the connection is good for

Connecting Ahrefs to ChatGPT is useful when we want to turn a repetitive SEO research task into a structured conversational request. For example:

1. Ask ChatGPT to retrieve the top 20 organic keywords for a competitor’s product category.
2. Request referring-domain trends for three competing pages.
3. Pull keyword gaps, sort by traffic potential, and group the results by topic.
4. Ask for a concise summary of changes before moving into the Ahrefs interface for validation.

Ahrefs’ current setup guide for ChatGPT Web instructs users to enable Developer Mode under advanced app settings, add the MCP connector, and use more structured prompts that specify the action, fields, and sorting required. That is a sensible workflow because vague prompts can produce vague analyses even when the underlying data is strong. (docs.ahrefs.com)

What it does not do

An MCP connection does not mean ChatGPT’s own buyer-facing answers will start citing your brand. It is a data-access connection for the user operating ChatGPT; it is not a direct influence channel into public ChatGPT retrieval, model behavior, or recommendation outputs.

That distinction matters. We can use Ahrefs MCP to investigate why a competitor ranks, earns links, or covers a topic. But to measure whether our company is visible in an actual AI answer, we still need to run the target prompt in the target engine, preserve the output, record cited URLs, and compare the result over time.

For teams deciding whether an integrated dashboard or independent measurement system is the better fit, our comparison of AI visibility tracker vs. cloud AI search visibility tools explains why ownership of raw prompts and evidence can matter as much as a polished report.

What Ahrefs AI citation charts measure—and what they do not

The October update added a citations chart to Brand Radar’s Overview. Ahrefs says Brand Radar distinguishes several visibility signals: mentions and citations can match entity name variations, while impressions and AI Share of Voice can match name variations or URLs attached to an entity. That makes the URL-tracking change especially relevant for publishers and brands with multiple product pages. (ahrefs.com)

A useful working definition is:

• Mention: The brand, product, organization, or entity appears in the generated answer.
• Citation: The answer identifies a source URL or domain associated with the entity or content.
• Impression: A tracked appearance opportunity in the reporting dataset.
• AI Share of Voice: A relative visibility measure within a selected prompt set or market, rather than a visit count.

These metrics are related but not interchangeable. A brand can be named in a shortlist without receiving a clickable source citation. A publisher can receive a citation without the company name featuring prominently in the prose. And an answer can include both a brand mention and a citation while still positioning a competitor as the preferred choice.

Ahrefs’ own citation research reinforces why this distinction is necessary. In its analysis of ChatGPT’s 1,000 most-cited pages from September 2025, Ahrefs found that only 32.3% were considered realistically outreach- or influence-worthy, while 28% had no organic Google visibility. Citation behavior therefore cannot be reduced to a standard ranking report. (ahrefs.com)

Why AI citations are not traffic, rankings, or conversions

A citation chart measures source visibility inside AI answers. It does not directly measure clicks, sessions, leads, pipeline, or revenue. We need separate systems for those outcomes.

Traditional search metrics answer questions such as: “Where do we rank for this keyword?” “How many clicks did Google send?” and “Which landing page converted?” AI answer metrics answer different questions: “Did the assistant recommend us?” “Did it cite our guide?” “Which competitor was named first?” and “What sources did it rely on?”

There may be overlap, but it is incomplete. Ahrefs’ research comparing citations across ChatGPT, Gemini, Copilot, and Perplexity with conventional results reported an average overlap of 11% between AI citations and Google/Bing top-10 rankings for the same query. That means a strong organic rank can help create a foundation, but it cannot serve as a proxy for citation visibility. (ahrefs.com)

Use a four-layer measurement model

We recommend keeping these layers separate:

This is the principle behind our guide to AI citations vs. AI visibility tracking: citations are an important evidence type, but they are only one part of whether a brand wins an AI-generated buyer answer.

Gemini, Copilot, Perplexity, and ChatGPT need separate baselines

Brand Radar’s October historical-data update gave ChatGPT and Perplexity up to five months of available chatbot history, with Gemini and Copilot at three months. That is helpful context, but the uneven history windows are a reminder not to compare every engine trend line as though it has identical coverage. (ahrefs.com)

Each engine can retrieve, format, cite, and recommend differently. Perplexity often foregrounds source links. ChatGPT responses can vary by product mode, browsing state, user context, and prompt wording. Gemini and Copilot can surface different source sets and answer structures for the same commercial question.

For that reason, we should build an engine-level baseline before declaring improvement or decline. For example, a B2B software company might track these five prompts across ChatGPT, Claude, Gemini, Perplexity, and Grok:

• “What are the best employee onboarding platforms for a 500-person company?”
• “Compare [our brand] with [competitor A] for IT onboarding.”
• “Which onboarding tools integrate with Okta and Workday?”
• “What is the best onboarding platform for distributed teams?”
• “How much does enterprise employee onboarding software cost?”

The point is not to chase a single aggregate score. It is to see where our brand is mentioned, where our domain is cited, which competitor owns the recommendation, and whether the answer changes after meaningful content, product, PR, or distribution work.

A practical prompt-level validation workflow

Vendor dashboards are useful starting points, but a repeatable evidence trail makes decisions more defensible. We use a prompt-level workflow because aggregate charts can hide the exact response that produced a change.

1. Define a stable prompt set

Start with 25 to 100 questions that map to actual buyer intent. Group them by funnel stage and use case: category discovery, comparison, alternatives, pricing, implementation, integrations, and problem solving.

Avoid prompts that are too broad, such as “best CRM.” A stronger prompt is “best CRM for a 50-person SaaS sales team using HubSpot and Slack.” It gives the AI engine a realistic decision context and makes competitive differences easier to interpret.

2. Capture the complete answer, not just a score

For every run, store:

• Exact prompt text
• Engine and model or mode where visible
• Run date and locale
• Full response text
• Brands mentioned and their order
• URLs cited, including destination domains
• Any qualification, caveat, or recommendation language

This turns an AI visibility observation into reviewable evidence. If a stakeholder asks why share of answer fell by 12 percentage points, we can point to the prompts where a competitor displaced us instead of guessing from an aggregate graph.

3. Classify the competitive result

For each answer, assign a simple outcome:

• Win: Our brand is explicitly recommended or placed first for the stated use case.
• Present: Our brand is mentioned but not clearly preferred.
• Cited only: Our site is cited but the brand is not meaningfully recommended.
• Lost: A competitor is recommended and we are absent.
• No decision: The response does not make a usable brand recommendation.

This gives us a practical definition of share of answer: the proportion of answers in which a brand earns a specified level of inclusion or recommendation. The exact rule should be documented before reporting begins.

Where URL tracking changes the measurement conversation

Name-only tracking can miss meaningful visibility. A brand may have a product subdomain, resource center, documentation domain, acquired brand, or flagship tool that appears in citations without being named in the prose.

Ahrefs’ October URL entity support addresses that gap by allowing URLs to be tracked alongside name-based entity mentions. For example, a company might track its company name, product name, primary domain, documentation subdomain, and a high-value research report URL under one measurement framework. (ahrefs.com)

That does not mean every domain citation should count as a brand win. A citation to a glossary page may provide credibility but not recommendation visibility. Conversely, an answer may name the brand repeatedly without linking to any owned URL. We should report both signals and preserve the distinction.

This is also where an independent, local-first approach can be useful. Our desktop tracker uses the customer’s own API key and keeps the individual prompt outputs that support the roll-up: the mention, cited URL, competitor, answer wording, and engine-specific result. That lets teams audit the underlying evidence rather than relying only on an opaque index.

For a broader framework, see our AI search measurement system, which separates visibility, citation, competitive, and business metrics into one operating model.

Does JSON-LD schema markup increase AI citations?

We should not treat JSON-LD schema markup as a direct AI-citation lever. Schema can make structured facts clearer for search engines and supports conventional rich-result and entity-understanding efforts, but there is no reliable public evidence that simply adding JSON-LD causes ChatGPT, Gemini, Copilot, or Perplexity to cite a page more often.

The mechanism matters. AI citations can be influenced by retrieval systems, source availability, freshness, answer format, domain reputation, factual usefulness, and engine-specific behavior. Schema may help machines interpret page attributes, but it does not create original evidence, establish authority, or guarantee selection as a source.

Ahrefs’ citation research found that AI-source behavior is not a mirror of organic visibility, and its analysis of 17 million citations reported that AI assistants tended to cite content that was 25.7% fresher than traditional organic results. That is directional research, not a universal rule, but it demonstrates why we should validate tactics against actual citations rather than assume a technical change worked. (ahrefs.com)

A better test is straightforward: choose a matched set of pages, make a documented schema change where appropriate, hold other major changes constant where possible, and monitor the same prompt set across multiple engines for several measurement cycles. If citations rise, inspect the answers and cited URLs before crediting the markup.

The other October Ahrefs features still matter

The AI-related additions received the most attention, but the other October changes can support the research workflow around them.

In Site Explorer, Ahrefs added entity filters in Organic Keywords, split branded traffic into a target brand versus other brands, added shared filter presets between Top Pages and Organic Keywords, and displayed URL Rating in Top Pages. Those updates can help us understand which conventional search topics and pages support a brand’s broader information footprint.

In Web Analytics, reports moved into sidebar navigation and page filters gained multi-URL pasting. In Google Search Console reports, Ahrefs added SERP overlays in the keywords table. And its API added a backlink-spam filter, Page Explorer endpoint, and Rank Tracker competitor metrics including Share of Voice, traffic, and position statistics. (ahrefs.com)

These capabilities are useful for diagnosing the environment around an AI visibility outcome. If a competitor suddenly dominates AI answers for integration questions, we might examine their supporting pages, entity coverage, links, ranking footprint, and content freshness. But conventional strength is evidence for investigation—not proof of why an AI engine selected that competitor.

Ahrefs reporting versus independent AI visibility tracking

Ahrefs is well positioned for teams that want SEO, competitive research, rank data, backlinks, analytics, and Brand Radar reporting in a broad platform. The October 2025 updates improved both the accessibility of Ahrefs data through MCP and the depth of its AI visibility reporting.

Our view is that reporting should be complemented by an independent validation layer when AI-answer visibility is a strategic metric. We need to know the exact prompt, engine, response, citation URLs, named competitors, and classification rule behind every trend line.

A local-first tracker is designed around that evidence. Rather than asking a team to trust a single vendor’s composite reporting method, it lets them run a defined prompt set using their own API key, save outputs locally, compare ChatGPT, Claude, Gemini, Perplexity, and Grok, and calculate share of answer from visible answer-level records.

The goal is not to replace Ahrefs. It is to prevent a common measurement mistake: treating an AI visibility dashboard as the final answer when it should often be the start of an investigation. The strongest operating model combines market-level reporting with prompt-level evidence and a clear distinction between mentions, citations, recommendation position, referral traffic, and conversion outcomes.

FAQ

How do you connect Ahrefs to ChatGPT?

Ahrefs’ current setup documentation directs users to enable Developer Mode in ChatGPT Web’s advanced app settings, add the Ahrefs MCP connector, authenticate the Ahrefs account, and use structured prompts. Access requires an eligible Ahrefs plan, and availability can depend on ChatGPT account features and Ahrefs plan limits. The MCP connection lets ChatGPT query Ahrefs data; it does not make ChatGPT cite your site. (docs.ahrefs.com)

What new AI visibility and citation features did Ahrefs release in October 2025?

The October 2025 roundup introduced hosted Ahrefs MCP access, up to five months of ChatGPT and Perplexity history, three months for Gemini and Copilot, URL entity tracking, and a Brand Radar citations chart. It also improved AI Responses and Topics reporting with charting, nested tables, saved filters, report names, and duplication. (ahrefs.com)

What do Ahrefs AI citation charts measure?

Ahrefs’ Brand Radar citations chart measures citation visibility within its tracked AI-answer dataset. It should be interpreted alongside mentions, impressions, and AI Share of Voice, not as a traffic or conversion report. A citation can indicate that an AI answer used or linked to a source, while a mention indicates that the brand appeared in answer text; either can occur without the other.

Can Ahrefs track brand mentions and citations across ChatGPT, Gemini, and Copilot?

Brand Radar’s October 2025 update included historical chatbot data for ChatGPT, Perplexity, Gemini, and Microsoft Copilot, plus support for tracking URLs and name-based entities. We should still evaluate each engine separately because coverage windows, source behavior, response formats, and recommendation logic differ. Cross-engine totals are useful summaries, but they can conceal meaningful engine-level gaps. (ahrefs.com)

Does adding schema markup increase AI citations?

Schema markup can help clarify structured information, but adding JSON-LD alone is not a proven way to increase AI citations. AI citation selection varies by engine and can involve retrieval, source quality, freshness, page usefulness, and query context. Treat schema as sound technical hygiene where it accurately represents the page, then test its impact through a fixed prompt set and captured citation evidence.

How do ChatGPT citations compare with traditional Google search traffic?

They measure different things. ChatGPT citations show that a page or domain was used as a visible source in an AI answer, while Google search traffic measures visits from traditional search results. A cited page may receive few clicks, and a high-ranking Google page may never be cited. Keep citation, referral, and conversion metrics separate before making investment decisions. (ahrefs.com)