AI Citations vs Backlinks: What Actually Builds Authority for SaaS and DTC Brands

Backlinks still matter, but mainly as an indirect signal. They help your pages rank, get crawled and get discovered, which feeds the pool of sources AI engines can draw from. ChatGPT citations also seem to depend on corroborating mentions on third-party sites, on content that is easy to extract, and on consistent facts about your product. 🦉

I should be candid about the evidence base. No AI engine publishes a full trust formula, and the research is young. Throughout this guide I label claims as documented (stated by the platform or shown in a linked study), observed (a correlation reported by researchers) or hypothesis (my own reasoning, which you should test). I link every study so you can check its date, sample and method yourself.

The short answer: authority now has two currencies

For roughly two decades, authority meant one thing to most SaaS and DTC marketers: links. These days I find it more useful to think in terms of two currencies.

• Link-based authority is earned through backlinks and mostly influences conventional search rankings.
• Mention-based authority is earned when credible sites refer to your brand consistently across the web. It plausibly influences whether an AI assistant names you when someone asks for a recommendation (hypothesis).

The two overlap, because a strong editorial placement often delivers both a link and a mention. They are not interchangeable, though. Take a query such as 'best project management tool for small teams'. A page that ranks well on the strength of its backlinks may be retrieved, but the final answer may still depend on which brands show up in reviews, comparison articles and community discussions, and on whether your own pages state facts clearly enough to be quoted.

How backlinks built authority in traditional SEO

The logic starts with PageRank. Google's founding research paper, The Anatomy of a Large-Scale Hypertextual Web Search Engine (Brin and Page, 1998), proposed using links between pages to estimate a page's importance, with each link counted as a vote weighted by the standing of the page casting it. Modern link analysis is far more sophisticated. Google's own documentation still lists PageRank among its ranking systems, and its Search Essentials note that links help it discover pages. Google is also clear that links are assessed alongside many other signals, so a link never guarantees a ranking boost.

The correlation evidence is substantial, with caveats:

• Backlinko (2020) analysed 11.8 million Google results and found the top result had, on average, 3.8 times more backlinks than results ranked second to tenth.
• Ahrefs (2020), studying roughly one billion pages, found that 66.31% had no backlinks and 90.63% received no organic search traffic from Google.

Both findings are observational. The first shows association, not that acquiring a set number of links causes a ranking gain. The second shows how unevenly links are distributed, not that unlinked pages cannot perform. Both are also several years old, so treat them as directional.

The standard SaaS and DTC playbook grew out of this: guest posts, resource-page outreach, affiliate and review links, and digital PR. In the UK, digital PR through national press and trade titles has been an especially productive lever, since a single placement can carry a high-authority link and real brand exposure at once.

The limitation that matters now is that a link is a proxy for trust. It does not describe what a brand is or does. A language model composing an answer needs to know what you sell, who it suits and how it compares, and a link graph says very little about any of that. Google also treats links created mainly to manipulate rankings, including paid links that pass ranking credit and excessive link exchanges, as link spam. The lesson that has held up is to earn editorially relevant links rather than chase raw volume.

How AI engines weigh sources differently for ChatGPT citations

Start with the distinction between two mechanisms, because they behave differently and reward different work.

• Parametric knowledge is what a model absorbed during training. It changes slowly and, in theory, reflects how consistently a brand was described across the web before the training cut-off.
• Retrieval-augmented answers use live search and citations. The engine fetches pages first and then writes an answer from them, so crawlable, current, easy-to-quote pages matter most.

For ChatGPT, OpenAI documents separate crawlers: OAI-SearchBot surfaces sites in ChatGPT search, while GPTBot is associated with model training. You can manage both through robots.txt. Allowing OAI-SearchBot makes you eligible to appear, but it does not guarantee a citation.

Here are the signals I consider most credible, each with its evidence status:

1. Entity consistency (hypothesis). This is how often and how coherently your brand appears alongside a category. It is an inference from how language models work, not a published criterion.
2. Third-party corroboration (hypothesis, partly observed). Reviews, comparison articles, forum discussions and industry publications are sources AI engines are often seen to surface, and agreement across independent sources makes a claim easier to verify. I have not found rigorous, public cross-engine data showing which domains are cited most or how much each type contributes, so I would not call this settled.
3. Content extractability (partly documented). Clear headings, direct answers, tables and clean HTML help. Google states there is no special markup required for AI features, but it recommends text-based content, internal linking and structured data that matches visible content.
4. Freshness (hypothesis). This is likely more relevant to engines that retrieve live results, such as Perplexity and Copilot, than to answers generated from training data.
5. Verifiable claims (observed, experimental). The GEO paper (Aggarwal et al., KDD 2024) reported that adding quotations, statistics and citations raised source visibility by up to 40% in its benchmark. That was a controlled experiment, not a forecast for commercial sites.
6. Platform differences (documented in principle). ChatGPT, Claude, Gemini, Copilot and Perplexity use different retrieval systems and source mixes, so any single-engine conclusion can mislead. Microsoft's Bing Webmaster Tools also offers an AI Performance report covering Copilot citations, a useful reminder that AI visibility can be measured separately from visits and rankings.

Part of why citation matters is that behaviour is shifting. A Pew Research Center analysis of browsing data from 900 US adults (68,879 Google searches, March 2025) found users clicked a traditional result in 8% of visits when an AI summary appeared, versus 15% without one. They clicked a link inside the summary in about 1% of visits, and about 18% of the searches observed displayed a summary. That is US Google behaviour, so UK patterns may differ, but being named in the answer is increasingly a separate outcome from earning the click.

Do backlinks still influence AI citations?

The honest position is mixed. Several industry studies report a positive relationship between domain authority or ranking position and being cited, especially for engines that lean on search indexes. These are correlations, and most come from vendors with their own methods and samples. Google does not say backlinks directly determine citation in AI Overviews, and OpenAI publishes no formula making backlinks a direct citation-ranking factor.

The indirect pathway is one plausible route: links improve rankings, rankings improve the odds a page is retrieved, and retrieval improves the odds it is cited. It applies mainly to search-grounded answers, where the engine fetches pages from a search index. It applies much less to answers written purely from a model's training data, where no page is retrieved at all. Even for search-grounded answers, each step is plausible and none is guaranteed.

The relationship also seems to break down in predictable places. A smaller or newer brand with strong third-party coverage can be recommended ahead of a higher-DR competitor, and some high-authority pages are never cited because their content is hard to extract or doesn't answer the question asked.

What about unlinked mentions? This is a hypothesis, not an established effect. Language models learn from text rather than only from link graphs, so a consistent unlinked mention in a credible context could strengthen the association between your brand and a category. Proving that would take controlled before-and-after tests. You could track citation rates for a fixed query set before and after a mention-only campaign, and compare them against a similar set of queries with no campaign. I have not seen published evidence of that kind, so I treat mentions as worth earning for their reputational value, not as a proven citation lever.

In practice, backlinks look like necessary-but-insufficient infrastructure. They support discovery and traditional authority, but they are not the whole strategy.

Backlinks vs AI citations: a side-by-side comparison

The table simplifies a messy picture. AI engines are not one system: a retrieval layer selects pages, and a language model then composes the answer. Engines differ, and much of the right-hand column reflects observation and inference rather than published specifications.

Dimension	Backlinks (link-based authority)	AI citations (mention-based authority)	Works in both?
What is measured	Votes of trust between pages	Whether you are cited or recommended in answers	Partly
Who decides	Search ranking systems, which Google documents (including PageRank)	Varies by engine: a retrieval layer chooses candidate pages, then a model writes the answer; the weighting is not published	No
Main signals	Quality and relevance of linking pages	Observed: corroborated mentions, extractable content, consistent facts	Editorial placements
Time to impact	Weeks to months	Variable: potentially faster for live retrieval, slower for training data	Both lag
How measured	Domain rating, referring domains, rankings	Query coverage, citations, share of voice, sentiment	Measure both separately
Unlinked brand mentions	Little direct value	Hypothesis: may strengthen entity association	Digital PR
On-page structure	Helps crawling and relevance	Makes passages easier to quote	Yes
Topical depth	Supports rankings	Supports category association	Yes
Review-site presence	Often nofollow or limited	Frequently seen as a repeated source for recommendations	Partly
Typical tactics	Guest posts, outreach, resource pages	Review profiles, expert commentary, community presence	Original research, expert commentary

The final column carries the practical message. Original research, digital PR and expert commentary can pay off in both worlds, which is why I advise shifting effort toward them rather than abandoning one currency for the other.

Should you stop link building entirely?

No. Abandoning links would undermine the search rankings that feed many retrieval systems, and organic traffic still drives revenue. Some context helps: a SparkToro and Datos study estimated that 58.5% of US and 59.7% of EU Google searches in 2024 ended without a click to the open web. Clicks are scarcer, but they haven't vanished. (The EU figure is the closest published proxy I know of for UK behaviour. It is not a UK-specific number.)

What I'd do instead is reframe link building as mention building. Prioritise placements on pages that AI engines already cite in your category, because a link on such a page serves rankings and, possibly, recommendation odds at once. Think of it as a portfolio shift rather than a prescribed split. Move a modest share of budget first, perhaps a quarter, and compare changes in AI visibility and referral traffic over a few months before moving more. I wouldn't claim any ratio is optimal, because nobody has the data to support one.

Avoid low-quality link schemes. Paid links that pass ranking credit fall under Google's link spam policy, and they add cost and risk without improving how AI engines describe you.

Where to shift your authority-building efforts for generative engine optimization

Google says there are no extra technical requirements for AI Overviews beyond regular Search eligibility, so I treat this as one discipline rather than a separate checklist. The steps below support traditional search and AI SEO alike, and I've flagged which ones are more speculative.

1. Audit the sources AI engines cite for your buying questions. Find where competitors appear and you don't. That gap list becomes your outreach target list. (This is a research method and assumes nothing about causation.)
2. Earn presence on those cited pages through digital PR, review platforms, partnerships and expert contributions. (This is standard good practice. The added citation benefit is a hypothesis.)
3. Check how legible your site is. Confirm crawlability, including OAI-SearchBot access in robots.txt and your server logs, and make sure structured data matches visible content, product descriptions are clear, and pricing and comparison pages are accessible. (Mostly ordinary technical SEO. The bot-specific access check is platform-specific and documented.)
4. Publish assets people can cite. Original data, benchmarks and definitive how-to content give others something to reference. HubSpot's State of Marketing and Shopify's Future of Commerce reports show the model, though publishing alone guarantees neither links nor citations.
5. Keep brand facts consistent across your site, directories, review profiles and social profiles. Contradictory pricing or feature claims weaken verifiability. (Good practice. The effect on AI answers is a hypothesis.)
6. Take part in communities such as Reddit, niche forums and LinkedIn, where genuine recommendations may influence what models learn and retrieve. Participate honestly. Manufactured enthusiasm is easy to spot and, for reviews, may breach UK consumer law, so check the rules before incentivising feedback.

None of this needs a separate toolkit. Each step is a content, technical or reputation task that also helps conventional search, and that is exactly why I trust these more than anything sold as an AI hack.

Special considerations for SaaS vs DTC brands

For SaaS brands:

• Publish comparison and alternatives pages, integration directories and quotable documentation.
• Maintain profiles on G2 and Capterra-style review platforms, which provide independent corroboration.
• Make claims verifiable by linking feature pages to documentation, explaining benchmark methodology and using named expert authorship. The Edelman and LinkedIn 2024 B2B thought leadership report found that 73% of surveyed decision-makers considered thought leadership a more trustworthy basis for assessing an organisation's capabilities than marketing materials, and 75% said it prompted them to research a product they had not previously considered. These are survey results, not proof of AI citation effects.

Here is an illustrative scenario (hypothetical, not a case study). A London-based HR software company finds that for 'best HR software for small UK businesses' the cited sources are two comparison articles and a review platform, and it appears on none of them. It would pitch a data-led piece to a UK trade title, complete its review-platform profile, and add an alternatives page with VAT-inclusive pricing in pounds. The outcomes to track are its citation rate for that query set, the number of cited sources that now mention it, and referral sessions from AI engines.

For DTC brands:

• Prioritise product reviews, ingredient or specification clarity, and retailer and marketplace listings.
• Keep product pages, product structured data and Merchant Center feeds aligned on price and availability. Eligibility for supported search experiences is not a guarantee of display.
• Earn placements in editorial gift guides and 'best of' lists, which resemble the way shoppers phrase their questions.

Another illustrative scenario (hypothetical): a UK skincare brand checks which review sites and 'best of' lists are cited for 'best fragrance-free moisturiser for sensitive skin'. It then makes sure its ingredient lists, pricing in pounds and delivery details are identical across its site, retailer listings and review profiles, and tracks whether its citation rate and position for that query set change over the next quarter.

For both:

Position within AI-generated lists plausibly shapes purchase-decision queries. Being named third rather than first may still count for something, but earlier placement is clearly preferable.

The UK angle:

UK publications, UK review sites such as Trustpilot and Which? (for DTC), UK trade press (for SaaS), and region-specific details like pounds pricing, UK delivery and VAT-inclusive prices could all influence what UK users see. I have not found UK-specific public evidence on this, so treat it as a hypothesis to test with a UK-located query set rather than an established ranking rule.

Measuring AI visibility and authority in an AI-first world

Domain rating and referring-domain counts describe your link graph, not your presence in answers. Measuring AI visibility needs its own method, and you can run it in a spreadsheet before you buy any tool.

A reproducible AI visibility testing protocol

1. Build a fixed query set. Use 20 to 50 buyer questions drawn from sales calls, support tickets, Search Console queries and review-site questions. Include category queries ('best X for Y'), comparison queries ('A vs B'), problem queries and alternatives queries. Freeze the list so results stay comparable.
2. Name your engines and test each separately, for example ChatGPT (with and without search), Gemini, Perplexity, Copilot and Claude. Never pool them, because their sources differ.
3. Standardise conditions. Use logged-out or clean sessions, a UK location and the same phrasing each time, and record the date and engine version where shown.
4. Repeat each query. AI answers are non-deterministic, so run each query at least three to five times per engine and record every run.
5. Record the same details every time. For each run, note whether you were mentioned, whether you were cited with a link, your position in any list, the sources cited, the competitors named and the sentiment (recommended, neutral or cautioned against).
6. Calculate against a baseline. Citation rate is appearances divided by total runs. Repeat on a fixed schedule, such as weekly or monthly, and compare against your baseline. Treat movements smaller than the run-to-run variation you observed as noise.
7. Test interventions fairly. To check whether a mention campaign works, compare targeted queries against a similar untouched set, and avoid changing several things at once.

The core metrics from this protocol are query coverage (the share of your questions where you appear at all), citation frequency and position, share of voice against competitors, sentiment, and the unlinked mentions that links-only reporting misses. Pair them with referral analytics, bearing in mind that, given the click behaviour Pew observed, citations will often outnumber visits.

Where an AI visibility tool fits

Manual testing gets laborious at scale, and that is where monitoring tools come in. At MentionOwl, we automate this protocol with daily monitoring across engines, competitor tracking and a 0-100 AI visibility score. To be clear, that score is a vendor-specific composite, not an industry standard. It combines query coverage, position-weighted citations, share of voice and soft mentions, so it is useful for tracking your own trend over time, but you shouldn't compare it with other tools' scores. Its weighting reflects our judgement rather than independent validation, so I recommend checking its movements against your own spot-test results.

A practical cadence:

1. Weekly: review movement in score, citations and competitor mentions.
2. Monthly: run a source-gap analysis against your target list.
3. Quarterly: adjust strategy and budget allocation.

What to do this quarter

For the first 30 days:

1. Baseline your AI visibility with the protocol above.
2. Identify the top ten cited sources in your category.
3. Fix legibility issues, starting with crawler access, product facts and comparison pages.
4. Start targeted outreach to the sources on that list, then re-test against your baseline.

If you would rather not run the baseline by hand, MentionOwl offers a 7-day trial (currently $1; check the site for pricing in your currency). It baselines your position, shows which competitors are being recommended and flags the legibility issues worth fixing first.

Frequently asked questions about ChatGPT citations and AI SEO

Do backlinks still matter if I'm optimizing for AI?

Yes, mostly indirectly. They support rankings, crawling and discovery, which influence what search-grounded AI answers can retrieve. They are not enough alone, since third-party mentions, content structure and factual consistency also appear to shape ChatGPT citations.

How do AI engines decide which sources to trust?

No engine publishes its full criteria. Observed patterns point to corroboration across reputable sources, topical relevance, extractable content, freshness for retrieval-based engines and consistent brand information, and each engine weighs these differently. Training-data answers and live-retrieval answers also work differently, so test them separately.

Should I stop link building entirely?

No. Shift emphasis toward placements on pages AI engines already cite in your category, and drop low-quality tactics that add risk without benefit.

How do I measure authority in AI search specifically?

Use a fixed set of buyer queries, test named engines separately on a schedule, repeat each query to allow for variation, and record citations, position, competitors and sentiment against a baseline. Tools such as MentionOwl can automate this, but the method works fine with a spreadsheet.

Do unlinked brand mentions help with AI visibility?

Possibly, but the evidence isn't there yet. Language models learn from text, so consistent mentions in credible contexts could strengthen how an engine associates your brand with a category. This remains a hypothesis until controlled tests show an effect, so treat mentions as worth earning for reputation while you test their impact on your own citation rate.