The LLM Consensus Playbook: What Ann Smarty Actually Runs for Clients

Most AI visibility advice is a vibe. Publish good content, be authoritative, build your brand. Nothing you can put on a Monday task list.

Ann Smarty runs an agency that sells this work, so her version is specific enough to scope. Twenty years in SEO, currently selling SEO audits and GEO as one product because they overlap too much to split, plus Reddit reputation work she fell into after fifteen years of being on the platform for free.

What follows is her sequence, pulled from the full Unscripted SEO interview and reorganized into the order you would actually execute it.

Step 1: Run a consensus audit before you run anything else

The organizing idea is what Ann calls LLM consensus.

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.

Not one great page. Many independent sources telling the same story. The audit is exactly what it sounds like: inventory everything on the open web that describes your company, then check whether those descriptions agree on who you are, what you sell, and how you differ from competitors.

Her honesty about her own footprint is the useful part for anyone scoping this. Twenty years of author bios, guest post byline blurbs, speaker pages and profile descriptions, all frozen at whatever was true when they were written. Some of her bylines still point at projects that have not existed for twenty years. She keeps fixing them and keeps finding more.

For how a single episode turns into links, articles and citations, see podcasting as content and SEO strategy.

Two practical notes she gave. First, the audit can be done in-house, it is time rather than tooling. Second, it scales with company age, so budget accordingly on a twenty year old business.

Do not skip your own site while you do it. Ann updated a personal site that was twenty two years old and untouched for five years, and saw LLM answers change. Detailed about page, real contact details, the trust signals. That is table stakes, not the strategy.

Step 2: Claim profiles, and yes, that includes directories

When asked what a brand new project should do first, her lead answer was the one nobody expected.

Do not dismiss directory submissions that easily these days, she said. LLMs use them heavily, and they have made a very good comeback from twenty years ago.

Directory submissions have been a punchline in this industry since roughly 2006. They are back on the list because LLMs read them.

The rest of the foundation layer:

• Name consistency across every profile you own
• Register where it makes sense for your niche. Yelp and Trustpilot for many businesses, industry-specific platforms for B2B
• Build a repeatable contribution system rather than a one-time blitz. Her framing: one asset can be a good update for ten profiles

This is the layer that matters most when there is nothing to audit yet. The question that prompted her answer came from a genuinely new build, a clean power project aimed at AI data centres, and for a business at that stage the consensus problem is inverted. There is no messy twenty year footprint to reconcile. There is no footprint at all, which means every source that gets created in the first year sets the story the models will repeat later. Get the profiles and the descriptions right once, at the start, and you never run the cleanup.

Then work out which channels drive LLM citations for your competitors where you are absent. Her usual list: Reddit, YouTube, Wikipedia, and a small set of high profile publications that are harder to crack but not impossible.

Step 3: Own the branded SERP before you spend a dollar outbound

This is the sequencing rule most teams get backwards, and it is the most quotable thing in the episode:

Being a brand is great, but no one talks about how you actually lose control over it, because stuff starts ranking for your name.

Growth generates branded search. Branded search is the query set you cannot afford to lose, and it is the one your competitors, review aggregators and unhappy Reddit threads will happily fill for you. If you have not done the foundation work, going outbound just funds faster discovery of whatever is already there.

The LLM angle makes it worse. Before a model recommends a brand in an AI answer, it runs the branded search, and it finds the things your customers were never meant to see. Ann has watched startups enjoy two good months and then watched the Reddit discussion start.

If your branded results are already a problem, that is a fixable project. We have written up the diagnostic path in when your brand disappears: fixing the brand name not ranking crisis.

Step 4: If you are doing Reddit, budget six months

Ann’s Reddit service is a long term, transparent, branded presence. Not thread hunting with links attached.

Her average is six months before the community really picks up, the subreddit starts ranking in Google, and it starts showing up in LLM citations. Once you have a working subreddit, branching into other communities gets easier.

The failure mode is not what you would guess. It is not new accounts getting flagged. She has watched very well established accounts, almost ten years old, get suspended on Reddit the moment they started trying to do something differently and quickly. Ten years of karma and history is not a shield. The moment an account turns promotional and starts linking out consistently, the clock starts.

Her reasoning for refusing shortcut Reddit work is worth repeating to any client who asks for it. Because they work on behalf of clients, they cannot afford the freedom and flexibility to game things and experiment. So they have to be more careful than anyone else.

She is explicit that she is not making a moral argument. People run the churn-and-burn model profitably. But when you operate a brand’s official presence, the blowback lands on the brand, and that is not a risk an agency gets to take on a client’s behalf.

Step 5: Audit what your own company says about itself

This is the deliverable nobody scopes and it is now a real line item.

We used to be an island, Ann said, describing the years when SEOs reported keyword movement and knew nothing else about the business. Now she sits with sales and customer support, because LLMs search spontaneously and will go find on Reddit whatever a company kept out of Google.

You would not believe, she said, how much of a reputation problem is caused by the gap between what the website says and what sales people promise. That gap causes an enormous amount of trouble on Reddit, and everywhere downstream of Reddit.

Pricing is her worked example. Undisclosed upfront costs surface on Reddit, then the AI overview states your upfront pricing back to the searcher. There is no version of this you win by staying quiet.

The support side of that gap is the half most SEO teams never touch, and it is the cheapest thing on this list to fix. Your reps are answering the exact questions the models are about to answer for you, usually with better information than your website carries, and none of it is written down anywhere a crawler can reach. Handing support the materials instead of hoping they improvise turns a liability into source content. Same answers, same language, now on a page.

So that alignment inside companies becomes the key to LLM visibility. And, in her words, it is a fight.

Build a plan B into the engagement, because you will lose this argument regularly. In fifty percent of cases she cannot win that battle, because the executives are still saying no, that is how we make money, we cannot say that.

When she loses it, she changes strategy instead of continuing to push. A lawyer client charges above market for a defensible reason, taking harder cases, and will not let her say so publicly. Her workaround is hosting AMAs and building authority around the constraint rather than through it.

That is a template. When the truth is off limits, stop optimizing the message and go build trust on a different surface.

Step 6: Put AI where it belongs in the workflow

If you assume the agency selling AI visibility is itself heavily automated, Ann’s answer will recalibrate you.

AI does: sentiment tracking, community monitoring, analytics, competitive research, ideation direction, internal linking suggestions, outreach and journalistic research.

Humans do: the actual Reddit posting, and the original ideas behind digital PR assets. They write manually there and do not automate the posting, because in her words it is not a good way to invent something actually original and unique. AI gives direction. Most ideation still comes from people brainstorming.

For entity audits, she connects Semrush or Ahrefs into Claude to surface angles she would not have reached alone. Then she spends three days to a week manually reading search results and sorting queries before writing a roadmap, because that is how she builds a real picture of a site’s positioning health.

Worth pairing with the data caveat: if you are pulling Google Search Console into any of this, remember that somewhere between thirty and eighty percent of the query table is simply missing. We covered why in Google Search Console data is not accurate. Automating analysis on top of an incomplete table just gets you to the wrong answer faster.

Step 7: Report on the answer, not the citation

Here is where Ann is harder on her own service line than most agencies would be in public.

Smarty Marketing reports LLM citations for a client’s official subreddit. She then explains why that metric is not the goal. Just one mention in a Reddit thread would not make you visible in AI answers. It can be a citation, but who cares about citations, they are not even clicked. The goal is for the brand to be in the answer.

One citation is not consensus. Consensus is many sources agreeing, which loops back to step one.

And her stance on the SEO-to-LLM relationship is refreshingly unhyped:

SEO is still the most predictable path into LLMs. Not the only one.

She put a number on the overlap in the same breath. Not one hundred percent, but thirty percent overlap is better than no overlap at all. Thirty percent, honestly stated, in a market full of people selling a hundred.

Her reason for anchoring on SEO anyway is a measurement argument, not a loyalty one. We cannot see prompts. Keyword data is the only available proxy for the questions being asked, and organic representation still correlates with showing up in AI answers. She will look into organic search anyway, whether a client asks her to or not, because there is more data there than anywhere else.

Which is notable given what walks through her door now. She has not had a lead in months who came to her saying they want to be on Google.

The one-page version

1. Audit your footprint for consensus. Everything that describes you should describe you the same way.
2. Fix your own site first. About page, contact details, trust factors. Table stakes.
3. Claim profiles. Directories included. Build a system so one asset updates ten places.
4. Own your branded results before you go outbound.
5. If Reddit is in scope, commit six months and stay transparent. Promotional turns kill old accounts.
6. Get sales and support in the room. Misalignment surfaces publicly and ends up in the AI answer.
7. Expect to lose that fight half the time and have a plan B ready.
8. Use AI for monitoring, research and direction. Keep originality and posting human.
9. Report on presence in the answer. Citations are not the win.
10. Pick one thing you can sustain weekly. Ann’s is a Wednesday live stream, and that alone drives a lot of business.

Listen to the full conversation

The full interview is on Unscripted SEO, hosted by Jeremy Rivera.

More from Ann: smarty.marketing · AnnSmarty.com newsletter and weekly live stream · her own write-up on LLM consensus · LinkedIn · @seosmarty

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