- Sarvesh Shrivastava argues that many SEO teams are getting weak AI output not because the model is limited, but because they start with thin context and no repeatable setup.
- His six-step Claude Fable 5 workflow treats business context, competitor data, model selection and buyer-intent filters as the foundation before any SEO prompt is written.
STOP using Claude Fable 5 like an amateur.
Most people type a question. Get an answer.
Move on.
Then wonder why they’re losing to competitors.
It’s not Claude’s fault.
It’s the setup.
Here’s the exact SEO setup I run before touching a single prompt:
1. Load your business… https://t.co/OJZWFvbV2N pic.twitter.com/YLj5UgO5cI
— Sarvesh Shrivastava (@bloggersarvesh) July 5, 2026
Sarvesh Shrivastava did not publish another prompt list.
On 5 July 2026, the Alventra Marketing founder and Favikon-ranked SEO creator shared a Claude Fable 5 setup framework that reads more like an operating process than a collection of AI tricks.
His argument is simple: most SEO practitioners do not lose because Claude gives bad answers. They lose because they ask from a weak setup.
The thread outlines six steps Shrivastava says he runs before touching a single prompt. The first is also the easiest to skip: load the model with full business context. That includes the company name, website, location, services, target cities and the top three competitor URLs.
His instruction for this step is direct: “Use this as context for everything. Never ask me for this again.”
That one line explains the whole framework. Shrivastava is trying to remove the generic starting point that makes so much AI-assisted SEO work sound interchangeable. Without business and competitor context, the model has to guess. With it, even short prompts can return sharper, more relevant answers.
The Setup Matters Before the Prompt
Shrivastava’s second step is model selection.
He tells users to open Claude Cowork, choose Opus 4.7 and turn on Extended Thinking. In his view, running serious SEO strategy through a default model setting lowers the quality ceiling before the work even starts.
Extended Thinking is designed to let Claude work through a problem before producing the final answer. For SEO tasks, that can matter because the model is not only rewriting copy or producing keyword ideas. It is being asked to reason through a market, a site, competitors and buyer intent.
The rest of Shrivastava’s framework follows the same logic. He recommends configuring global instructions inside Claude Cowork, creating a competitor file in markdown, loading that file before every audit session and applying a buyer-intent filter to keyword research.
The buyer-intent step is especially practical. Instead of generating long keyword lists filled with informational queries, Shrivastava pushes the model toward searches where someone is closer to calling, booking or buying. That cuts down the volume of low-value keyword ideas and keeps the work tied to commercial outcomes.
Why Fable 5 Raises the Bar for Context
Shrivastava’s thread arrived just days after Claude Fable 5 returned to global availability.
Anthropic said US export controls were applied to Claude Fable 5 and Claude Mythos 5 on June 12, forcing a temporary suspension while nationality verification systems were added. The controls were lifted on June 30, and by July 1 the model was available again across Claude Platform, Claude.ai, Claude Code and Claude Cowork.
The model also returned with additional safeguards. Anthropic said it retrained a safety classifier to address a bypass method flagged in an Amazon research report, with the updated system blocking the technique in more than 99% of cases.
For SEO teams, the export-control story is less important than what the model can do now that it is back in use. Anthropic describes Claude Fable 5 as a Mythos-class model with leading scores across SWE-Bench Pro, FrontierCode Diamond and Humanity’s Last Exam.
That is where Shrivastava’s critique becomes stronger. A more capable model does not remove the need for clear context. It makes poor setup more wasteful.
If the model can reason more deeply, but the user only gives it a vague request, the output still starts from a weak brief. The benchmark score does not fix missing business context, unclear positioning or a competitor set the model has never seen.
Thin Context Still Produces Thin Output
Shrivastava’s competitor file requirement is one of the more useful parts of the framework.
Instead of asking Claude to “analyze competitors” from scratch every time, he builds a reusable markdown file and loads it before audit work. That gives the model a stable reference point for positioning, content gaps, keyword overlap and local market comparison.
This connects with a broader shift in SEO. As AI systems become more involved in search, content selection and citation behavior, practitioners need to understand both the retrieval layer and the prompt layer. Our analysis of AI citation signals and SEO visibility shows how quickly that technical layer is changing.
But prompt quality still matters. A strong model with weak context will still produce broad recommendations. A properly briefed model can make smaller prompts work harder because the foundation is already loaded.
That is the main lesson from Shrivastava’s setup. The prompt is not the starting point. The context is.
Cost Makes Weak Setup More Expensive
There is also a practical cost argument.
Claude Fable 5 is priced at $10 per million input tokens and $50 per million output tokens, which makes careless usage more expensive than it was with cheaper models.
That does not mean every SEO task needs the highest-capability model. Many routine tasks can still run on cheaper options. But when teams use Fable 5 for audits, strategy, competitive analysis or high-value content planning, the input quality matters more.
A vague request wastes tokens and produces output that still needs heavy correction. A structured setup uses the model’s reasoning power on the actual problem.
That is why Shrivastava treats setup as part of the work, not preparation before the work. Business context, competitor files, global instructions and buyer-intent rules all shape the quality of the final answer.
The Model Still Has Limits
Fable 5 is not a frictionless upgrade for every task.
The model uses a two-stage classifier for sensitive areas including cybersecurity, biology and model distillation. Flagged queries can be routed to Opus 4.8, with users notified when that happens.
For most SEO workflows, that will not be a major issue. Local SEO audits, content planning, keyword filtering and competitor analysis are unlikely to trigger those systems.
The more relevant trade-off is speed. Extended Thinking can slow the workflow down. For quick rewrites or basic metadata, that may not be worth it. For strategy work that affects months of content, rankings or client direction, the slower response is usually easier to justify.
What SEO Teams Should Take From This
None of Shrivastava’s six steps is revolutionary on its own.
Business context, competitor mapping, global instructions and buyer-intent filters have all existed in prompt engineering for some time. What his thread makes clear is that many SEO practitioners still treat them as optional upgrades instead of the baseline.
That is the real gap.
A team using Claude Fable 5 with persistent business context, a clean competitor file and clear commercial filters is not doing the same work as someone typing one-off prompts into a blank chat. Even if both use the same model, they are not operating from the same starting point.
The wider SEO landscape is moving in the same direction. AI systems are changing how content is created, retrieved and evaluated. Prompting is only one part of that shift, but it is still the part practitioners can control immediately.
Shrivastava’s framework is useful because it brings the work back to basics. Before asking for an audit, strategy or keyword plan, the model needs to know the business, the market, the competitors and the type of demand that matters.
Claude Fable 5 may be more capable than earlier models, but it does not magically understand a business it has not been briefed on. The teams that build that briefing process into their workflow will get more from the model than those still treating AI like a smarter search box.
