SEO SOP: Running an LLM Consensus Audit

This SEO SOP gives agencies, in-house teams and freelancers a repeatable process for auditing every source on the open web that describes a business, finding where those sources contradict each other, and making them agree. It distills the central play from an Unscripted SEO interview with Ann Smarty of Smarty Marketing, and belongs to our library of repeatable SEO SOPs.

Objective

Watch: Ann Smarty on LLM consensus, Reddit and brand search.

The objective of this SOP is to produce a single, consistent story about a business across every source a language model can reach, so that when an LLM assembles an answer about the brand it finds many independent sources saying the same thing. Ann Smarty calls this state LLM consensus. The deliverable is a scored inventory of every describing source, a canonical description to measure them against, and a prioritized fix list. This is the step that runs before profile building, before outbound spend, and before any Reddit or community work.

I refer to that as LLM consensus. This is when LLMs are able to find a lot of different sources about you and they are very consistent in their messaging. Don’t be like me. My footprint is a mess.

Ann Smarty, Smarty Marketing

Key Steps

1. Fix your own website first. Before auditing anything external, make sure the site is current: a genuinely detailed about page, real contact details, and the rest of the trust factors. Ann updated a personal site that was twenty two years old and largely untouched for five years, and saw LLM answers change as a result. Treat this as table stakes rather than as the strategy.
2. Write the canonical description. Produce one short reference paragraph covering what the business does, its value proposition, and how it differs from competitors. Every other source gets measured against this. Ann names those three things specifically as what has to be consistent, so do not skip the differentiation line just because it is the hard one to write.
3. Inventory every source that describes the business. Sweep the open web for anything carrying a description: author bios and guest post bylines, directory and review listings, social and platform profiles, speaker and event pages, partner and vendor pages, press coverage, and old profiles from previous eras of the company. Ann’s own example is that some of her bylines still describe projects that have not existed for twenty years. Capture the URL, the description it carries, and who controls it.
4. Score each source against the canonical description. Mark every row as agrees, contradicts, or stale. Contradictions are the priority, not gaps. A source that describes a business you no longer are does more damage to consensus than a source that says nothing, because the model has to reconcile two competing claims and may surface the wrong one.
5. Fix in order of reachability. Work the list in three passes: profiles you can edit right now, profiles that need credential recovery, and pages that need someone else to make the change for you. (inferred) The credential recovery pass is usually the real bottleneck, and it is worth scoping separately because a five year old account with a dead email attached can take longer than everything else combined.
6. Close the internal gap between marketing, sales and support. Interview the sales and support teams and compare what they tell prospects against what the website says. Ann is explicit that a large share of reputation damage comes from that mismatch, and that it surfaces on Reddit and then everywhere downstream. Undisclosed upfront pricing is her worked example: if the number is not published, people find it on Reddit and the AI overview repeats it back to the searcher anyway.
7. Map the competitor citation gap. Identify the channels that are driving LLM citations for competitors and where the business has no presence at all. Ann’s usual list is Reddit, YouTube, Wikipedia, and a small set of high profile publications that are harder to crack but still reachable. Do not dismiss directory submissions here. She is direct that LLMs make heavy use of them and that they have made a real comeback.
8. Build the contribution system and set a re-audit cadence. Rather than a one-time blitz, build a repeatable way to push updates outward, on the principle that one asset can be a good update for ten profiles. Then put the audit back on the calendar, because the footprint keeps growing and the fix list is never actually empty.

Cautionary Notes

• Scope by company age, not company size. Ann is explicit that the audit takes a lot of time and that older businesses take the longest. A twenty year old company has twenty years of accumulated descriptions to reconcile. Quote and staff accordingly.
• It is never finished. Ann still finds stale references to her own dead projects and describes the cleanup as never ending. Sell it as a maintained state, not a one-off project that closes.
• Expect to lose the internal alignment argument about half the time. Ann puts it at fifty percent of cases where she cannot win the battle, because executives will not publish the thing that they believe is how they make money. Have a plan B in the engagement rather than continuing to push. Her workaround for a lawyer client who could not explain their above-market pricing was hosting AMAs and building authority around the constraint instead of through it.
• One citation is not consensus. Ann’s own warning about her flagship service applies here too: a single mention in a single thread will not make a brand visible in AI answers, and citations are not even clicked. The goal is presence in the answer, which requires many sources agreeing.
• Do not treat the audit as a replacement for the website. Ann would not neglect the owned site even in a distributed footprint model. The site is where the canonical description lives.
• Do the audit before outbound spend, not after. Growth generates branded search, and branded search is where uncontrolled results surface. Running ads into an unaudited footprint just funds faster discovery of whatever is already sitting there.

Tips for Efficiency

• Run it in-house. Ann says directly that an LLM consensus audit can be done in-house. The constraint is time, not tooling or licence cost, which makes it one of the few high-leverage plays a small team can execute without buying anything.
• Recycle one asset across ten profiles. Repackaging existing website content into profile updates is what turns the fix list from a grind into a system.
• Use AI for direction and research, not for the reading. Ann connects Semrush or Ahrefs into Claude to surface competitive angles she would not have reached alone, but still spends three days to a week manually reading actual search results and sorting queries before writing a roadmap. Automating that step gets you a faster wrong answer.
• Start from organic search data. Prompts are not observable, so keyword and organic data remain the best available proxy for the questions people are actually asking. Ann says she looks at organic search whether the client asks her to or not, because there is more data there than anywhere else.
• Pair the audit with one sustainable weekly habit. Ann cannot afford polished video and her team is full, so her single non-negotiable is a live stream every Wednesday, and she credits that consistency alone with driving a lot of business. Pick the bare minimum you can sustain and do it every week.

Sources & Relevant Episodes

Ann Smarty, Smarty Marketing. Twenty years in SEO, now selling SEO audits and GEO as a single product because they overlap too much to separate, plus Reddit reputation work. She framed LLM consensus as the very first step for any business that cares about being visible in AI answers, and was candid that her own twenty year footprint is the cautionary example rather than the model.

Full write-up of the conversation: The LLM Consensus Playbook: What Ann Smarty Actually Runs for Clients. The episode itself is at Unscripted SEO, hosted by Jeremy Rivera. Ann’s own write-up on the idea is on her newsletter.

Related episodes on the same problem: Grant Simmons on entity SEO · Jason Barnard on entity optimization and AI · Jon Henshaw on AI hype and sources of truth