Do Analyst Firms Affect What AI Recommends?

Do assistants cite analyst firms?

One of them does very well and the rest are invisible, which is not the picture the category's self-description implies.

No assistant operator publishes citation shares by domain, so this rests on vendor observation. The largest published dataset is Profound's, released 5 June 2025 and updated August 2025, covering 680 million citations across ChatGPT, Google AI Overviews and Perplexity from August 2024 to June 2025. Profound sells answer-engine optimisation software and has a clear commercial interest in the finding that AI citations concentrate on a short list of influenceable domains; the figures are reported with that stake named.

Gartner is Perplexity's third most-cited domain at 1.0% of all citations, behind only Reddit at 6.6% and YouTube at 2.0%, and Google AI Overviews' fifth at 0.7%. It does not appear in ChatGPT's top ten at all — where the B2B slot is filled by G2 at 1.1%. Forrester, IDC, Omdia, Everest Group, ISG and the rest appear in no platform's published top ten.

Two readings of this are worth separating. The flattering one for analyst relations is that a single paid-subscription research firm has become one of the three most-consulted sources on an entire assistant. The less flattering one is that the effect is confined to Gartner and to two of three platforms, that even there it is 1% of citations rather than 10%, and that a share of all citations across an undisclosed prompt mix says nothing about the share on the prompts a software buyer actually types. The same denominator problem runs through which sources AI assistants cite.

There is one non-vendor confirmation that the mechanism is real, and it comes from a regulatory filing rather than a study. Gartner's annual report on Form 10-K for 2025, filed 12 February 2026, carries this risk factor: “some of our content is exposed to the datasets leveraged by AI chatbots, and these chatbots may provide substantive content, either with or without contribution, in query responses to users which could reduce the need to enter our websites”. The same filing notes that search engines and LLMs “often update their proprietary algorithms, which affects the placement of links to our websites”. Gartner is telling investors, under securities liability, that assistants are surfacing its material and that this is a risk to its traffic. Forrester's 2026-filed 10-K approaches the same subject from the other side, warning that third parties “may load our proprietary research into large language models in violation of our terms of use, which could reduce the value of our services”.

What an assistant can actually read

This is the part that changes how the rest of the question should be framed, and it can be checked directly rather than taken on anyone's word.

We read the three firms' robots files on 26 September 2026. Forrester's names eight AI and dataset crawlers individually — GPTBot, ClaudeBot, Claude-User, Claude-SearchBot, Google-Extended, CCBot, FacebookBot and Omgilibot — and gives each of them the same single instruction: Disallow: /report/*. Nothing else on the site is closed to them. IDC's robots file consists of a wildcard user-agent with an empty disallow, which permits everything and names no AI crawler at all. Gartner's robots file could not be retrieved: every request, with and without a browser user-agent string, returned an automated-access challenge page rather than the file.

Forrester's configuration is the informative one because it is deliberate. Somebody chose those eight agent names and chose to fence exactly one path. The effect is that an assistant can read Forrester's blogs, press releases, policy pages and marketing copy, and cannot read the reports. The published research — the Waves, the Landscapes, the Tech Tides — is the material a buyer would most want an assistant to have consulted, and it is the one directory that is explicitly withheld. The wider pattern of who names which crawler is in which AI crawlers publishers name, and what a block does and does not accomplish is in does blocking a crawler remove you from search.

Now put that beside the licensing rules, because together they produce a consequence nobody advertises. Forrester's Research Usage Policy states that clients “cannot download and share Forrester reports with non-licensed individuals or third parties”, that distributing PDFs or printed copies is prohibited, and that a downloaded PDF is restricted to the individual who obtained it and must be deleted when the licence expires. The one sanctioned route to a publicly readable copy is an External Reprint licence, which “enable[s] clients to distribute full reproductions of published reports on websites or in physical form”, with customisation available for the introduction and specific sections.

So the analyst report that an AI crawler can actually reach is, in the ordinary case, not the one on the analyst's own site. It is the reprint a vendor has paid to license and hosted on its own domain, sometimes with a vendor-written introduction attached. The corpus available to an assistant is therefore selected by which vendors bought reprint rights, which is a different selection from what the analysts concluded. That is a structural bias with no bad actor in it, and it is the single most important thing in this guide. The general problem of research that sits behind a licence is the subject of does paywalled coverage get indexed, and the training-data dimension is in can you tell an AI not to train on your site.

What the money actually is

Both Gartner and Forrester are listed companies and file audited annual reports, which makes this the one part of the analyst business that can be described from primary documents rather than from rumour.

Gartner reported total revenue of $6.5 billion for 2025, of which the Insights segment produced $5.07 billion at a 77% gross contribution margin. Subscription products were approximately 78% of total revenue. Insights contract value stood at $5.155 billion, split between Global Technology Sales at $3.91 billion and Global Business Sales at $1.245 billion. The firm describes being “in steady contact with over 13,000 distinct client enterprises worldwide”, with more than 2,400 experts and more than 510,000 direct client interactions in 2025, sold by 3,704 quota-bearing GTS sales associates and 1,280 in GBS. In August 2025 it launched AskGartner, an AI tool for licensed users. Net income fell to $0.7 billion from $1.3 billion, principally on a goodwill impairment.

The definition attached to the largest of those numbers is where the pay-to-participate question quietly goes unanswered. Global Technology Sales, per the 10-K, “sells products and services to users and providers of technology”. Buyers and vendors are reported in one figure. There is no public breakdown of how much of the $3.91 billion comes from the companies Gartner rates, and that is the number anyone assessing the firm's independence would most want. It is not a disclosure failure — segment reporting is not designed to answer this — but it means the most common claim in both directions, that vendor money dominates or that it is marginal, is unverifiable from the filings.

Forrester is a much smaller and currently shrinking business, and treating it as Gartner's peer overstates it. Total 2025 revenue was $396.9 million, down 8%: Research $295.6 million, Consulting $88.2 million, Events $13.1 million. Goodwill impairment equalled 27.8% of revenue and the net loss 30.1%; in February 2026 the company announced it would stop selling strategy consulting engagements. Gartner is roughly sixteen times Forrester's size by revenue.

Two things in the smaller firm's accounts matter more than its size. Forrester's revenue description names the product directly: it derives revenue from “subscriptions to our Research products and services, subscriptions to, and individual licenses of, electronic ‘reprints’ of our Research, performing consulting projects and advisory services, and hosting events”. And in the third quarter of 2024 it launched a subscription reprint product; in 2025, while revenue from subscriptions to its research declined, “an increase in reprint revenue” partly offset it. The line that grew is the line a vendor buys in order to republish a favourable rating.

IDC cannot be described this way. It files no US public-company reports — a full-text search of SEC filings turns up other companies citing IDC data, not IDC's own accounts — so there is no audited statement of how its revenue divides between buyers and vendors, and we are not going to characterise it from secondary reporting.

What a vendor can and cannot buy

Forrester publishes this in unusual detail, and because Gartner's pages were unreachable, Forrester is where the documentary answer comes from. It should not be assumed to describe every firm.

What is not for sale, in the firm's own words

Forrester's Integrity Policy states that “client companies cannot purchase research coverage or favorable opinions”. Analysts are restricted from joining vendor boards or endorsing particular products or strategies. Board members “are restricted from being personally involved in any evaluative research processes”. Commissioned studies and consulting deliverables intended for a client's external use “must be company-, product-, and brand-agnostic”, with one named exception: the Total Economic Impact deliverable, which is by design about a single client's product. Analysts may have reasonable travel expenses reimbursed for vendor events and briefings, which is disclosed rather than hidden.

The Wave methodology adds the enforcement detail. The analyst defines the scope and the vendor set; Forrester “reserves the right to evaluate products and services, regardless of vendor participation” and will not remove a vendor on request, though it may remove one it judges incompatible with the evaluation's objectives. Since July 2024 vendors are placed in three bands — Leaders, Strong Performers, Contenders — using a three-point scoring rubric with criteria weighted by importance. Analysts decline client engagements with vendors they are evaluating in the same market. The report carries the framing that it “is a guide for buyers considering their purchasing options in a technology marketplace and is based on our analysis and opinion”.

The non-participation policy is the clearest statement of where the leverage actually sits. A vendor that declines to take part is still evaluated, from public information, secondary research, independent client interviews, reference feedback and estimates. What it forfeits is procedural: it is not covered by the Wave Vendor Review Policy, and therefore cannot see its scorecard and write-up before publication, cannot submit feedback, and cannot escalate. Participation does not buy a score. It buys a proof stage.

What that proof stage is, precisely

A participating vendor receives its own scorecard and its own write-up, including where it lands on the graphic. It may give “a limited amount of feedback”. It may escalate factual inaccuracies, first to the Research Director and then to the Group Director and a Wave Center of Excellence. It may not dispute criteria, scales or weightings, which Forrester calls “a matter of opinion”; may not dispute Forrester's interpretation of customer feedback; may not raise concerns about how other vendors scored; and may not ask Forrester to depart from its own process. Escalations are “not automatically accepted”.

What is for sale

The right to say it, in public, at scale. Forrester's Citation Policy limits non-clients to citing public intellectual property — blogs and press releases — plus their own participation in a study and quotes attributed to themselves. Any external or commercial use of the research requires express written permission from a Citations department, reachable by a named email address and phone number. Three licensing pathways exist: research licences, reprint licences and commissioned research contracts, each carrying different citation privileges and costs.

The prohibitions are as informative as the permissions. A vendor may not reproduce full research documents without a licence, may not create analyst quotes or attribute statements to researchers, may not use Forrester data to make negative comparisons against competitors, may not put Forrester branding in a press-release header or an email signature, and may not claim endorsement. Citations must carry attribution and, for certain research types, an objectivity statement clarifying that Forrester does not endorse any company on the basis of its ratings.

Read as a whole, the structure is coherent and it is not the structure most commentary assumes. There is no price list for a quadrant position. There is a price list for distribution, and it is enforced by copyright rather than by editorial discretion. A vendor's analyst budget mostly buys subscription access, advisory time, the proof stage, and the licence to republish — and it is that last item that determines what ends up on a public, crawlable URL. The same distinction between earning a rating and renting the right to advertise it runs through what business awards cost and the disclosure rules in what the FTC requires for paid placement.

Peer Insights and the review corpus

Gartner Peer Insights is a distinct asset from Gartner's research, and the ownership picture around it is settled and easy to get wrong. Our guide to whether review sites affect what AI recommends establishes that G2's marketplace network now includes Capterra, Software Advice and GetApp — the properties formerly traded as Gartner Digital Markets — while Peer Insights remained with Gartner, and sets out the platforms' gating, verification and incentive rules. We are not going to restate any of it here.

What belongs in this guide is one observation and one refusal. The observation is the platform asymmetry. ChatGPT's top ten contains G2 at 1.1% and no Gartner; Perplexity's and Google's contain Gartner at 1.0% and 0.7%. If the assistants were converging on one authoritative B2B source they would be reaching for the same one, and they are not. A vendor who optimises for one and infers the other has a measurement problem rather than a strategy.

The refusal is about Peer Insights' own rules. Gartner publishes review policies, submission criteria, incentive rules and vendor review-solicitation programmes, and we could not open any of those pages. Every attempt against gartner.com on 26 September 2026 — the Peer Insights pages, the Magic Quadrant methodology page and the robots file, tried directly and through a text-extraction proxy — returned either an automated-access challenge or a sign-in wall. We are not going to describe a methodology we have not read, so Peer Insights appears in this guide as context rather than as a source of rules. Anyone quoting specific Peer Insights criteria should be asked where they read them.

One adjacent piece of structure is worth naming because it is easy to miss. Omdia, the analyst firm, sits inside Informa TechTarget alongside more than thirty B2B trade publications and a purchase-intent data business; the company's own 10-K says Omdia is positioned to serve clients “on both the buy-side and sell-side of the technology market”. A vendor can therefore buy editorial adjacency, intent data about the readers, and analyst coverage from one counterparty. We work through that filing in do trade publications outperform national outlets.

Does any of this move a shortlist or an answer?

Here the evidence runs out, and it is better to say so than to bridge the gap with an inference.

What is established: Gartner is reached for heavily by two of three assistants; Gartner itself says in a filing that chatbots are surfacing its content; the reports are mostly closed to AI crawlers at source while vendor-hosted reprints are not; and the right to produce those reprints is a priced product whose revenue is growing at one of the two listed firms.

What is not established: that improving a vendor's analyst standing changes what an assistant says about it. We could find no controlled study — vendor, academic or platform — that varies a company's Wave band, Magic Quadrant position or Peer Insights rating and measures the resulting change in a “best X for Y” response. The mechanics of how those answers are assembled are in how AI assistants recommend companies and what signals answer engines use, and neither closes this gap.

What can be measured is something narrower and more honest: how often companies invoke analyst ratings in documents they are legally accountable for. We ran a full-text search of SEC EDGAR for Form 10-K filings made between 1 January 2025 and 26 September 2026. “Gartner” appears in 137, “Magic Quadrant” in 17, “Forrester Wave” in 11, “IDC MarketScape” in 10 and “Peer Insights” in 7.

Three things follow for a vendor. Treat the reprint, not the rating, as the visibility asset. If the report is disallowed to eight named AI crawlers and the licensed reprint on your own domain is not, then the reprint is the version an assistant can read, and whether you licensed it is the variable you actually control.

Price the proof stage honestly. Participation in a Wave buys the right to see your scorecard, correct a factual error and escalate once. It does not buy a band, and declining does not buy absence. Anyone quoting a fee for a position is describing something other than the published process.

Do not pay for analyst relations on an AI-visibility argument alone. The argument requires a causal step from a 1% citation share to a changed recommendation, and nobody has published that step. Pay for analyst coverage because procurement committees and boards ask for it, which is a documented, ordinary reason, discussed in does media coverage affect B2B buying and third-party authority for SaaS.

Media Morpheus sells media and podcast placement, not analyst relations, and nothing here would support adding it. The finding that should reach a founder is narrower than a tactic: one analyst firm is genuinely central to two assistants, the research those firms sell is largely closed to the crawlers those assistants use, the copy that is open is the one a vendor paid to license, and the causal claim attached to all of it has never been tested.

Methodology and limitations

Citation shares are from “AI Platform Citation Patterns: How ChatGPT, Google AI Overviews, and Perplexity Source Information” by Nick Lafferty, Profound, published 5 June 2025 and updated August 2025, covering 680 million citations observed August 2024 to June 2025. Profound sells answer-engine optimisation software. The study reports only each platform's top ten domains, so Forrester, IDC, Omdia and Gartner-in-ChatGPT are below the reporting threshold and their shares are unknown rather than zero; the prompt set is not published, so the query mix cannot be matched to a software buyer's questions; and the products change continuously, so the shares are a period observation. Financial figures are from two primary regulatory filings: Gartner, Inc.'s annual report on Form 10-K for the year ended 31 December 2025, filed with the SEC on 12 February 2026, and Forrester Research, Inc.'s Form 10-K for the same period, filed 13 March 2026. Contract value is not revenue and the two are not interchangeable; Gartner's Global Technology Sales figure combines technology users and technology providers in a single number, so the share of Gartner's income that comes from the companies it rates cannot be derived from public filings and is not estimated here. IDC files no US public-company reports, so no comparable audited figures exist and we have not characterised its revenue mix from secondary reporting. Policy statements are quoted from Forrester's own published policy pages — the Integrity Policy, the Citation Policy, the Research Usage Policy, the Forrester Wave Methodology, the Wave Vendor Review Policy and the Wave Vendor Participation (non-participation) policy — all read 26 September 2026. These are the publisher's assertions about its own conduct: Forrester publishes no audit, enforcement log or exception rate against them, so their effectiveness is not verifiable from outside, and Forrester has a commercial interest in being regarded as independent. Crawler rules were read directly from each firm's robots file on 26 September 2026; a robots directive is a request that well-behaved crawlers honour and is not a technical block, and a file can change at any time. The SEC EDGAR phrase counts were run by us on 26 September 2026 over Form 10-K filings dated 1 January 2025 to 26 September 2026; they count filings containing a phrase and therefore include self-references, competitor references and citations of market-size data, so they are an upper bound on vendors promoting their own ratings and are not evidence about AI behaviour. Three limits are decisive. First, and most seriously, Gartner's own documentation was unreadable: every request to gartner.com on 26 September 2026 — Peer Insights policy pages, the Magic Quadrant methodology page and the robots file, attempted directly and through a text-extraction proxy — returned an automated-access challenge or a sign-in wall, so no Gartner methodology, review rule or inclusion criterion is quoted anywhere above, and the documentary picture in this guide is Forrester's and should not be assumed to describe Gartner or IDC. Second, we relied on our companion guide for the ownership and gating facts about Peer Insights and the review platforms rather than re-verifying them. Third, we found no study of any kind that manipulates a vendor's analyst standing and measures the resulting change in an AI recommendation; the central causal question here is unanswered in the public record and we have left it unanswered rather than inferred.

Conclusion

Partly, unevenly, and through a route almost nobody describes. One firm — Gartner — is genuinely among the handful of domains that Perplexity and Google AI Overviews reach for, at around 1% of citations, and is missing from ChatGPT's top ten entirely; Forrester, IDC and Omdia are below the reporting threshold everywhere. Gartner itself told investors that chatbots are surfacing its content and may reduce the need to visit its site. But the reports themselves are largely fenced off: Forrester disallows eight named AI crawlers from its report directory while leaving everything else open, and the only sanctioned way a full report reaches a public URL is an External Reprint licence bought by a vendor. So the analyst corpus an assistant can read is, in the main, the vendor-selected, vendor-hosted subset. As for what money buys: not a rating — Forrester states flatly that coverage and favourable opinions are not purchasable, and evaluates vendors whether they participate or not — but access, advisory time, a pre-publication proof stage, and the licence to republish, which is the item that determines visibility. The question a founder wants answered, whether a better analyst position produces a better AI recommendation, has not been tested by anyone who published the answer. On present evidence the defensible move is to be correctly scoped into the evaluations that matter, to participate for the proof stage rather than for a score, to license the reprint if public visibility is the goal, and to disbelieve anyone who quotes a citation share as though it were a causal effect.

Frequently Asked Questions

Sources

1. regulatory filing Gartner, Inc., 2026. Annual Report on Form 10-K for the fiscal year ended December 31, 2025. US Securities and Exchange Commission, filed 12 February 2026. Source for total revenue of $6.5 billion, Insights revenue of $5,072.6 million, Insights contract value of $5,155 million, the GTS/GBS split and its “users and providers of technology” definition, client and analyst counts, the AskGartner launch and the AI-chatbot risk factor.
2. regulatory filing Forrester Research, Inc., 2026. Annual Report on Form 10-K for the fiscal year ended December 31, 2025. US Securities and Exchange Commission, filed 13 March 2026. Source for total revenue of $396.9 million and its segment split, the goodwill impairment and net loss, the reprint revenue line and the subscription reprint product launched in Q3 2024, and the large-language-model risk factor.
3. vendor Lafferty, N., 2025. AI Platform Citation Patterns: How ChatGPT, Google AI Overviews, and Perplexity Source Information. Profound, 5 June 2025, updated August 2025. 680 million citations, August 2024 to June 2025; top ten domains per platform only. Profound sells answer-engine optimisation software.
4. firm policy Forrester. Integrity Policy. Read 26 September 2026. “Client companies cannot purchase research coverage or favorable opinions”; the brand-agnostic rule for externally used commissioned work and the Total Economic Impact exception; analyst and board restrictions.
5. firm policy Forrester. Citation Policy. Read 26 September 2026. Non-client limits, the express-written-permission requirement, the three licensing pathways, and the prohibitions on fabricated analyst quotes, negative comparisons, branding in press-release headers and claims of endorsement.
6. firm policy Forrester. Research Usage Policy. Read 26 September 2026. Sharing limits for licensed clients, the prohibition on distributing reports to non-licensed third parties, and the Internal and External Reprint licences that permit full reproduction on a client website.
7. firm policy Forrester. Forrester Wave Methodology. Read 26 September 2026. Scope and vendor selection, the right to evaluate regardless of participation, the three bands adopted in July 2024, the three-point rubric and weighted criteria.
8. firm policy Forrester. Forrester Wave Vendor Review Policy. Read 26 September 2026. What a participating vendor sees before publication, the two-level escalation route, and the categories of dispute Forrester declines.
9. firm policy Forrester. Forrester Wave Vendor Participation Policy. Read 26 September 2026. Non-participants are evaluated from public information, secondary research, client interviews, reference feedback and estimates, and lose review and escalation rights.
10. crawler policy forrester.com/robots.txt and idc.com/robots.txt. Read 26 September 2026. Forrester names GPTBot, ClaudeBot, Claude-User, Claude-SearchBot, Google-Extended, CCBot, FacebookBot and Omgilibot and disallows each from /report/*; IDC's file blocks nothing and names no AI crawler. Gartner's file could not be retrieved.
11. regulatory filing TechTarget, Inc. (Informa TechTarget), 2026. Annual Report on Form 10-K for the fiscal year ended December 31, 2025. US Securities and Exchange Commission, filed 11 March 2026. Source for Omdia sitting alongside the trade-media and purchase-intent businesses and for the “buy-side and sell-side” description.
12. official records US Securities and Exchange Commission. EDGAR full-text search. Form 10-K filings, 1 January 2025 to 26 September 2026, queried 26 September 2026: “Gartner” 137 filings, “Magic Quadrant” 17, “Forrester Wave” 11, “IDC MarketScape” 10, “Peer Insights” 7.